MCP-USE integration with inclusion of context7 mcp server integration #211
Conversation
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the You can disable this status message by setting the WalkthroughThis update introduces comprehensive Model Context Protocol (MCP) integration into the RA.Aid system. It adds new dependencies and configuration options for MCP-Use and Tree-sitter servers, updates the CLI with flags for MCP configuration and server management, and implements a synchronous MCP client wrapper for tool integration. Prompt templates and agent logic are extended to dynamically load and inject server-specific guidance based on active MCP servers. Documentation is updated to detail new dependencies, configuration, and usage patterns. Several new guidance files and a default MCP server configuration are included to support these features. Changes
Sequence Diagram(s)sequenceDiagram
participant User
participant CLI (main)
participant MCPUseClientSync
participant ToolLoader
participant Agent
participant PromptSystem
User->>CLI (main): Launch with MCP flags (e.g., --mcp-use-config)
CLI (main)->>MCPUseClientSync: Initialize with config (load/merge/disable servers)
MCPUseClientSync->>MCPUseClientSync: Start event loop, create sessions, load tools
CLI (main)->>ToolLoader: get_custom_tools(mcp_use_client)
ToolLoader->>MCPUseClientSync: get_tools_sync()
ToolLoader-->>CLI (main): Combined custom and MCP tools
CLI (main)->>Agent: Start agent (e.g., planning/research)
Agent->>PromptSystem: Build prompt (inject {mcp_guidance})
PromptSystem->>Agent: Prompt with loaded MCP guidance
Agent->>CLI (main): Run agent logic using tools and guidance
CLI (main)->>MCPUseClientSync: Cleanup on exit
Possibly related PRs
Poem
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
There was a problem hiding this comment.
Actionable comments posted: 5
🧹 Nitpick comments (5)
ra_aid/tool_configs.py (1)
58-58: Function signature updated appropriately to support MCP-Use tools.The updated signature allows for passing an optional
MCPUseClientSyncinstance, enabling integration with MCP-Use tools while maintaining backward compatibility.Consider updating the function docstring to document the new
mcp_use_clientparameter:def get_custom_tools(mcp_use_client: Optional[MCPUseClientSync] = None) -> List[BaseTool]: """Dynamically import and return custom tools from the configured module. The custom tools module must export a 'tools' attribute that is a list of langchain Tool objects (e.g. StructuredTool or other tool classes). Tools must return a Dict with keys: - success: bool - can_retry: bool - return_code: int - output: str If can_retry=True, the tool may be retried with the previous output appended to the prompt, up to max_retries times. + Args: + mcp_use_client: Optional MCPUseClientSync instance to load MCP-Use tools from + Returns: List[BaseTool]: List of custom tools, or empty list if no custom tools configured """ra_aid/utils/mcp_use_client.py (2)
47-50: Return a copy to protect internal state
get_tools_sync()returnsself._toolsdirectly.
Callers can mutate the list and inadvertently corrupt the wrapper’s cache.- return self._tools + # Return a shallow copy so external code cannot mutate internal list + return list(self._tools)
69-80: Error propagation is good – but shield against partial initialisationIf
create_all_sessions()oradapter.create_tools()fails,_clientmay be
Nonewhile some resources (e.g. the process that runs the MCP server) are already
started. Make sure to wrap the async body in atry/exceptand perform
await client.close_all_sessions()on failure to avoid orphaned processes.ra_aid/__main__.py (2)
939-954:atexitcleanup is great – add manual shutdown on fatal errorsIf the main program raises before normal interpreter shutdown, the
atexithandler
still fires, but only after stack unwinding. To be safer, expose the instance in
config_repo(or a dedicated singleton) so that other fatal-error handlers can call
mcp_use_client_instance.close()explicitly beforesys.exit(1).No diff needed; just a recommendation.
955-957: Pass tool list instead of client?
get_custom_tools(mcp_use_client=mcp_use_client_instance)couples the function to the
client implementation. Consider passing the already-materialised list of tools
(mcp_use_client_instance.get_tools_sync()) to decouple concerns and enable easier
testing / mocking.
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (6)
examples/context7.json(1 hunks)pyproject.toml(1 hunks)ra_aid/__main__.py(5 hunks)ra_aid/prompts/research_prompts.py(1 hunks)ra_aid/tool_configs.py(3 hunks)ra_aid/utils/mcp_use_client.py(1 hunks)
🧰 Additional context used
🧬 Code Graph Analysis (1)
ra_aid/tool_configs.py (1)
ra_aid/utils/mcp_use_client.py (2)
MCPUseClientSync(10-80)get_tools_sync(47-49)
🔇 Additional comments (6)
pyproject.toml (1)
61-61: LGTM! MCP-Use dependency added correctly.The addition of the
mcp-use>=1.2.7dependency aligns with the PR objective of integrating MCP-USE repository. The version constraint allows for compatibility with future updates.ra_aid/prompts/research_prompts.py (1)
114-115: Excellent prompt guidance for Context7 tool prioritization.The added guidance effectively instructs the agent to prioritize using
resolve-library-idandget-library-docstools when working with specific libraries or frameworks. This aligns perfectly with the PR objective of creating a more extensible agent that can access up-to-date documentation.ra_aid/tool_configs.py (2)
39-39: LGTM! Required import for MCP-Use integration.Properly imports the
MCPUseClientSyncwrapper class that's needed for the MCP-Use integration.
111-124: Well-implemented MCP-Use tools integration with proper error handling.The implementation correctly:
- Attempts to get tools from the MCP-Use client only when provided
- Extends the existing tools list with the MCP tools
- Provides clear console feedback with appropriate color coding
- Handles exceptions during tool retrieval
This approach ensures that the MCP-Use tools are properly integrated while maintaining robustness through appropriate error handling.
ra_aid/__main__.py (2)
86-87: Unused import?
atexitis only used after ~939. ✔️ Nothing to change.
275-278: Epilog examples read well – but reflect new mutual-exclusion flagThe comment line advertises
--mcp-use-config … --no-context7, which is correct.
No action needed.
examples/context7.json
Outdated
| { | ||
| "mcpServers": { | ||
| "context7": { | ||
| "command": "npx", | ||
| "args": ["-y", "@upstash/context7-mcp@latest"] | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
💡 Verification agent
🧩 Analysis chain
Well-structured configuration for Context7 MCP server integration.
This configuration file properly defines the Context7 MCP server using npx to install and run the latest version of the package. This enables the agent to access up-to-date documentation from thousands of packages as mentioned in the PR description.
Consider the implications of using @latest tag:
🏁 Script executed:
#!/bin/bash
# Description: Check if the context7-mcp package exists and get its version info
echo "Checking if @upstash/context7-mcp package exists on npm registry..."
npm view @upstash/context7-mcp version descriptionLength of output: 255
Lock Context7 MCP server version for stability
The @upstash/context7-mcp package exists on npm (current latest: v1.0.6). While npx @upstash/context7-mcp@latest always pulls the newest build, it can introduce breaking changes unexpectedly. To ensure reproducible runs, consider pinning to a specific version or a semver range.
• File: examples/context7.json
- Change
"@upstash/context7-mcp@latest"
to
"@upstash/context7-mcp@1.0.6"
or
"@upstash/context7-mcp@^1.0.6"
Adds Task Master to the default MCP config, renames config file and flag, and updates planning prompt to suggest using Task Master tools.
- Integrates Task Master tools via default MCP config.\n- Adds --disable-task-master-planning flag to control planning behavior.\n- Updates Planner agent prompt to conditionally use Task Master workflow.\n- Makes MCP client loading more robust.\n- Addresses coderabbitai feedback from previous PR (race condition, close(), paths, validation).\n- Includes examples/ directory in package data.
- Adds mcp-server-tree-sitter dependency and to default MCP config.\n- Implements automatic project registration with the server on startup.\n- Updates Research agent prompt to encourage using Tree-sitter tools.
- Adds granular MCP disabling via --disable-default-mcp [servers...].\n- Adds warning for missing Task Master API keys.\n- Adds hint about Tree-sitter build dependencies on MCP init error.\n- Implements Task Master state synchronization by passing task_master_id through implementation tool and prompting Planner to call set-status.
- MCP client now tracks successfully initialized servers.\n- Agent prompts only include guidance for active MCP servers.\n- MCP initialization errors are handled more gracefully, disabling only failed servers.\n- Refactors MCP config loading logic in __main__.py.
- Adds GitHub MCP server to default config.\n- Adds Docker prerequisite and GITHUB_TOKEN note to README.\n- Adds check for GITHUB_TOKEN in __main__.py.\n- Updates Planner agent prompt with GitHub tool guidance.
- Creates ra_aid/prompts/mcp_guidance/ directory with .txt files for each server.\n- Modifies planning and research agents to load guidance from these files based on active servers.\n- Updates prompt templates to use a single {mcp_guidance} placeholder.\n- Ensures guidance files are included in package distribution via pyproject.toml.
- Fix JSON syntax in default MCP config.\n- Fix include paths in pyproject.toml.\n- Add missing Python imports.\n- Remove unreachable code in tools/agent.py.\n- Add timeouts to MCP client thread wait/join.\n- Correct grammar in Tree-sitter guidance.
- Add timeouts to MCP client thread wait/join.\n- Add missing type hint import in __main__.py.\n- Refactor tree-sitter registration logic slightly.
feat: Comprehensive MCP Server Integrations
Package version issue
|
Wow, this should be a huge improvement if we can get this running properly. Great work so far. Lmk if any help is needed. |
|
unfortunately, the "mcp use" package works only in python 3.11 so you'd have to create a venv to use this as it is now. I know there's already mcp capabilities in ra-aid. But if this is still something that is liked, maybe its ok to review. In my forked repo I've added taskmaster mcp to handle the planning for the agent (when taskmaster mcp is activated) as well as a few other mcp servers that have needs and am now updating it to be an interactive loop that expands beyond seeing one task to completion. I also added github mcp, so the agent can use those tools to manage a repo. But I don't know if that's beyond the scope of what's desired currently All that to say, any help is welcome |
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (6)
ra_aid/utils/mcp_use_client.py (1)
15-28: Fix docstring formatting issueThe class docstring has a formatting issue where the "Note:" paragraph (lines 22-26) appears to be in the middle of the main description, with line 27 continuing from line 22.
"""Synchronous wrapper around ``mcp_use.MCPClient``. ``mcp_use`` is fully async; RA.Aid's tool-pipeline expects plain synchronous ``langchain_core.tools.BaseTool`` instances. This wrapper starts a background event-loop, initializes the MCP client and converts every server tool to a ``StructuredTool`` via the official + ``LangChainAdapter``. + Note: Using MCP servers often requires external dependencies managed by that server. For example, using the Context7 server requires ``npx`` (Node.js) to be available in the environment to run ``@upstash/context7-mcp``. - ``LangChainAdapter``. """README.md (3)
251-253: CLI arguments formatting inconsistencyThe formatting of these CLI arguments doesn't match the pattern used for other arguments in the document. The loose punctuation marks (as flagged by static analysis) make the list harder to parse visually.
Consider reformatting these entries to match the style of other CLI arguments in the document:
- - `--mcp-use-config`: Path to a custom MCP-Use JSON config file. If provided without `--disable-default-mcp`, only this config will be loaded. - - `--disable-default-mcp [SERVER ...]`: Disable specific default MCP servers (context7, taskmaster-ai, tree_sitter, github) or all if no names are given. Can be used with `--mcp-use-config` to merge custom config with filtered defaults. - - `--disable-task-master-planning`: Force RA.Aid to use its internal planner instead of Task Master tools, even if the Task Master MCP server is active. + - `--mcp-use-config`: Path to a custom MCP-Use JSON config file. If provided without `--disable-default-mcp`, only this config will be loaded. + - `--disable-default-mcp [SERVER ...]`: Disable specific default MCP servers (context7, taskmaster-ai, tree_sitter, github) or all if no names are given. This can be used with `--mcp-use-config` to merge custom config with filtered defaults. + - `--disable-task-master-planning`: Force RA.Aid to use its internal planner instead of Task Master tools, even if the Task Master MCP server is active.🧰 Tools
🪛 LanguageTool
[uncategorized] ~251-~251: Loose punctuation mark.
Context: .... ---disable-default-mcp [SERVER ...]: Disable specific default MCP servers (c...(UNLIKELY_OPENING_PUNCTUATION)
[style] ~251-~251: To form a complete sentence, be sure to include a subject.
Context: ..., github) or all if no names are given. Can be used with--mcp-use-configto merg...(MISSING_IT_THERE)
[uncategorized] ~252-~252: Loose punctuation mark.
Context: ...ults. ---disable-task-master-planning: Force RA.Aid to use its internal planne...(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~253-~253: Loose punctuation mark.
Context: ... server is active. ---expert-provider: The LLM provider to use for expert know...(UNLIKELY_OPENING_PUNCTUATION)
157-158: Node.js capitalization issueThere's a minor capitalization issue with "Node.js" as flagged by the static analyzer.
- **Node.js & npm/npx:** Required for the default MCP servers (Context7, Task Master). Install from [nodejs.org](https://nodejs.org/). + **Node.js & npm/npx:** Required for the default MCP servers (Context7, Task Master). Install from [Node.js](https://nodejs.org/).🧰 Tools
🪛 LanguageTool
[uncategorized] ~157-~157: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...(NODE_JS)
159-162: Indentation inconsistency in list itemsThe static analyzer flagged indentation inconsistencies in the unordered list items. The OS-specific instructions are indented with 4 spaces instead of the standard 2 spaces for Markdown.
- **C/C++ Build Tools:** Required by the Tree-sitter MCP server (`mcp-server-tree-sitter`) to build language parsers. Installation varies by OS: - - **Debian/Ubuntu:** `sudo apt-get update && sudo apt-get install build-essential` - - **Fedora:** `sudo dnf groupinstall "Development Tools"` - - **macOS:** Install Xcode Command Line Tools: `xcode-select --install` - - **Windows:** Install Visual Studio with C++ build tools (select "Desktop development with C++" workload in the installer). + **C/C++ Build Tools:** Required by the Tree-sitter MCP server (`mcp-server-tree-sitter`) to build language parsers. Installation varies by OS: + - **Debian/Ubuntu:** `sudo apt-get update && sudo apt-get install build-essential` + - **Fedora:** `sudo dnf groupinstall "Development Tools"` + - **macOS:** Install Xcode Command Line Tools: `xcode-select --install` + - **Windows:** Install Visual Studio with C++ build tools (select "Desktop development with C++" workload in the installer).🧰 Tools
🪛 markdownlint-cli2 (0.17.2)
159-159: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
160-160: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
161-161: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
162-162: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
ra_aid/__main__.py (2)
1022-1022: Unnecessary f-string placeholderThere's an unnecessary f-string without placeholders as flagged by the static analyzer.
- logger.info(f"Attempting to initialize MCP-Use client...") + logger.info("Attempting to initialize MCP-Use client...")🧰 Tools
🪛 Ruff (0.8.2)
1022-1022: f-string without any placeholders
Remove extraneous
fprefix(F541)
1045-1053: Nested if statements can be simplifiedThe static analyzer flagged two instances of nested if statements that can be simplified.
- if task_master_planning_enabled: - if not os.getenv("ANTHROPIC_API_KEY"): - logger.warning("Task Master planning is enabled, but ANTHROPIC_API_KEY environment variable is not set. Task Master tools requiring it may fail.") + if task_master_planning_enabled and not os.getenv("ANTHROPIC_API_KEY"): + logger.warning("Task Master planning is enabled, but ANTHROPIC_API_KEY environment variable is not set. Task Master tools requiring it may fail.") - if "github" in active_mcp_servers: - if not os.getenv("GITHUB_TOKEN"): - logger.warning("GitHub MCP server is active, but GITHUB_TOKEN environment variable is not set. GitHub tools will likely fail.") + if "github" in active_mcp_servers and not os.getenv("GITHUB_TOKEN"): + logger.warning("GitHub MCP server is active, but GITHUB_TOKEN environment variable is not set. GitHub tools will likely fail.")🧰 Tools
🪛 Ruff (0.8.2)
1045-1046: Use a single
ifstatement instead of nestedifstatements(SIM102)
1051-1052: Use a single
ifstatement instead of nestedifstatements(SIM102)
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (14)
README.md(3 hunks)examples/default_mcp_servers.json(1 hunks)pyproject.toml(2 hunks)ra_aid/__main__.py(6 hunks)ra_aid/agents/planning_agent.py(2 hunks)ra_aid/agents/research_agent.py(3 hunks)ra_aid/prompts/mcp_guidance/context7.txt(1 hunks)ra_aid/prompts/mcp_guidance/github.txt(1 hunks)ra_aid/prompts/mcp_guidance/taskmaster-ai.txt(1 hunks)ra_aid/prompts/mcp_guidance/tree_sitter.txt(1 hunks)ra_aid/prompts/planning_prompts.py(1 hunks)ra_aid/prompts/research_prompts.py(1 hunks)ra_aid/tools/agent.py(3 hunks)ra_aid/utils/mcp_use_client.py(1 hunks)
✅ Files skipped from review due to trivial changes (6)
- ra_aid/prompts/mcp_guidance/github.txt
- ra_aid/prompts/mcp_guidance/context7.txt
- ra_aid/prompts/planning_prompts.py
- ra_aid/prompts/mcp_guidance/tree_sitter.txt
- ra_aid/prompts/mcp_guidance/taskmaster-ai.txt
- examples/default_mcp_servers.json
🚧 Files skipped from review as they are similar to previous changes (2)
- ra_aid/prompts/research_prompts.py
- pyproject.toml
🧰 Additional context used
🧬 Code Graph Analysis (2)
ra_aid/tools/agent.py (1)
ra_aid/agent_context.py (2)
reset_completion_flags(54-58)reset_completion_flags(195-199)
ra_aid/agents/planning_agent.py (1)
ra_aid/database/repositories/config_repository.py (1)
get_config_repository(178-194)
🪛 LanguageTool
README.md
[uncategorized] ~157-~157: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...
(NODE_JS)
[uncategorized] ~251-~251: Loose punctuation mark.
Context: .... - --disable-default-mcp [SERVER ...]: Disable specific default MCP servers (c...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~251-~251: To form a complete sentence, be sure to include a subject.
Context: ..., github) or all if no names are given. Can be used with --mcp-use-config to merg...
(MISSING_IT_THERE)
[uncategorized] ~252-~252: Loose punctuation mark.
Context: ...ults. - --disable-task-master-planning: Force RA.Aid to use its internal planne...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~309-~309: Consider an alternative adjective to strengthen your wording.
Context: ... Abstract Syntax Trees (ASTs), enabling deeper understanding of code relationships, sy...
(DEEP_PROFOUND)
[uncategorized] ~315-~315: Loose punctuation mark.
Context: ...able-default-mcp context7 taskmaster-ai: Disable specific default servers. - --...
(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~316-~316: Loose punctuation mark.
Context: ...s. - --mcp-use-config ./my_config.json: Use only a custom MCP configuration. - ...
(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~317-~317: Loose punctuation mark.
Context: ...tion. - --disable-task-master-planning: Keep Task Master tools available but us...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~319-~319: This wording can make your sentence hard to follow. Try rephrasing for improved clarity.
Context: ...ult MCP server fails to initialize (e.g., due to missing dependencies), it will be autom...
(DUE_TO_BECAUSE)
🪛 markdownlint-cli2 (0.17.2)
README.md
159-159: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
160-160: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
161-161: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
162-162: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
🪛 Ruff (0.8.2)
ra_aid/__main__.py
1022-1022: f-string without any placeholders
Remove extraneous f prefix
(F541)
1045-1046: Use a single if statement instead of nested if statements
(SIM102)
1051-1052: Use a single if statement instead of nested if statements
(SIM102)
🔇 Additional comments (26)
ra_aid/utils/mcp_use_client.py (5)
30-56: LGTM! Proper thread synchronization implementedThe initializer correctly avoids the race condition mentioned in past reviews by using a threading event with timeout to ensure the event loop is running before scheduling coroutines. The error handling is also robust with appropriate logging.
68-98: Excellent improvements to the close() methodThe close() method has been significantly improved from past reviews:
- Added idempotency with the
_closedflag- Proper checking if the loop is running before scheduling
- Added timeout handling for session closing
- Added thread aliveness check after join with appropriate warning
These changes prevent resource leaks and make shutdown more robust.
103-111: LGTM! Proper event loop managementThe _run_loop method correctly sets up the event loop, signals when it's ready via
_loop_started.set(), and ensures the loop gets closed when finished.
126-148: LGTM! Robust error handling per serverThe code properly handles initialization errors on a per-server basis, allowing some servers to fail without affecting others. Clear, detailed logging helps with troubleshooting.
169-202: LGTM! Well-implemented tree-sitter auto-registrationThe auto-registration for tree-sitter is carefully implemented with proper error handling and verification at each step. The code correctly uses await for the async tool call.
ra_aid/agents/research_agent.py (3)
19-21: LGTM! Clean importThe added import is properly placed with related imports and uses the conventional alias for importlib.resources.
372-389: LGTM! Well-implemented dynamic MCP guidance loadingThis implementation correctly:
- Retrieves active MCP servers from configuration
- Skips Task Master guidance in the Research agent (handled separately in Planning agent)
- Uses proper error handling with appropriate logging levels for different error cases
- Efficiently concatenates guidance parts into a single string
417-417: LGTM! Properly integrates MCP guidanceThe combined MCP guidance is correctly integrated into the research prompt template.
ra_aid/agents/planning_agent.py (3)
19-20: LGTM! Clean importThe added import is properly placed with related imports and uses the conventional alias for importlib.resources.
341-368: LGTM! Well-implemented task master-aware guidance loadingThis implementation properly:
- Handles Task Master guidance conditionally based on a separate flag
- Processes other servers based on activity status
- Uses appropriate error handling with different logging levels
- Has clear, logical organization
The conditional handling of taskmaster-ai is a good design choice that allows for more flexibility.
376-376: LGTM! Properly integrates MCP guidanceThe combined MCP guidance is correctly integrated into the planning prompt template.
ra_aid/tools/agent.py (3)
3-3: LGTM! Appropriate import updateThe import line is properly updated to include Optional for the new parameter type.
372-384: LGTM! Well-documented new parameter and return typeThe function signature and docstring have been properly updated to include the new task_master_id parameter with clear documentation on its purpose and the updated return structure.
466-475: LGTM! Improved return structureThe updated return structure with a clean dictionary is more modular and better supports the MCP integration by preserving the task_master_id for upstream processing.
README.md (3)
153-169: Well-structured documentation of new dependenciesThe documentation clearly explains the additional dependencies required for MCP integration, with specific OS installation instructions. This is excellent for user experience.
🧰 Tools
🪛 LanguageTool
[uncategorized] ~157-~157: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...(NODE_JS)
🪛 markdownlint-cli2 (0.17.2)
159-159: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
160-160: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
161-161: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
162-162: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
172-175: Good practice for API key requirements documentationClear documentation of required API keys is essential for users to successfully use the MCP integration features.
303-319: Comprehensive MCP integration documentationThe documentation provides clear explanations of the MCP server capabilities, benefits, and how to manage them through command-line flags. The note about automatic disabling on dependency failures is especially helpful.
🧰 Tools
🪛 LanguageTool
[style] ~309-~309: Consider an alternative adjective to strengthen your wording.
Context: ... Abstract Syntax Trees (ASTs), enabling deeper understanding of code relationships, sy...(DEEP_PROFOUND)
[uncategorized] ~315-~315: Loose punctuation mark.
Context: ...able-default-mcp context7 taskmaster-ai: Disable specific default servers. ---...(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~316-~316: Loose punctuation mark.
Context: ...s. ---mcp-use-config ./my_config.json: Use only a custom MCP configuration. - ...(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~317-~317: Loose punctuation mark.
Context: ...tion. ---disable-task-master-planning: Keep Task Master tools available but us...(UNLIKELY_OPENING_PUNCTUATION)
[style] ~319-~319: This wording can make your sentence hard to follow. Try rephrasing for improved clarity.
Context: ...ult MCP server fails to initialize (e.g., due to missing dependencies), it will be autom...(DUE_TO_BECAUSE)
ra_aid/__main__.py (9)
512-528: Well-structured CLI argument additions for MCP integrationThe new CLI arguments are well-defined with appropriate help text that aligns with the README documentation. The metavar for SERVER_NAME and default=None are good choices for the implementation.
535-544: Excellent file validation improvementAdding eager validation for file paths provided via arguments is a significant improvement that will prevent confusing errors later in execution. This addresses issues like the one mentioned in past review comments where paths were only validated when they were used.
934-1011: Robust MCP configuration management logicThe implementation handles multiple configuration scenarios elegantly:
- User config only when no defaults are disabled
- Filtered defaults when user disables specific servers
- Merged configs when both user config and defaults are needed
Error handling is comprehensive with appropriate fallbacks. The comments explaining the different cases make the code more maintainable.
1015-1035: Well-implemented MCP client initialization with cleanupThe code effectively initializes the MCP client and properly registers cleanup with atexit. The error handling ensures the application continues to function even if MCP initialization fails.
🧰 Tools
🪛 Ruff (0.8.2)
1022-1022: f-string without any placeholders
Remove extraneous
fprefix(F541)
1038-1054: API key validation for MCP serversGood implementation of environment variable checks for required API keys, with appropriate warning messages that explain the potential impact.
🧰 Tools
🪛 Ruff (0.8.2)
1045-1046: Use a single
ifstatement instead of nestedifstatements(SIM102)
1051-1052: Use a single
ifstatement instead of nestedifstatements(SIM102)
1056-1056: Proper integration with existing systemsThe update to the
get_custom_tools()call properly integrates the MCP client with the existing tool system, enabling the MCP tools to be made available to the agent.
91-93: Good use of standardized modulesUsing
importlib.resourcesfor accessing package resources andatexitfor resource cleanup follows Python best practices and ensures proper resource management.
5-10: Proper type annotations for improved code qualityThe addition of proper type annotations using TYPE_CHECKING ensures type checking without runtime overhead, following good Python development practices.
282-284: Helpful examples in CLI help textThe addition of MCP-related examples in the CLI help text is useful for users to understand how to use the new features, matching the documentation in the README.
| ### Prerequisites | ||
|
|
||
| - **Python 3.10+** | ||
| - **Git** | ||
| - **Ripgrep (`rg`)**: Required for efficient code searching. Install via your system package manager (e.g., `apt install ripgrep`, `brew install ripgrep`, `choco install ripgrep`). | ||
|
|
||
| - **Docker:** Required if using the default GitHub MCP server integration (which runs in a container). Install from [docker.com](https://www.docker.com/products/docker-desktop/). Ensure the Docker daemon is running. |
There was a problem hiding this comment.
Upgraded Python requirement may impact compatibility
The project now requires Python 3.10+ (upgraded from 3.8+). This change should be highlighted prominently to users since it represents a potentially breaking change that could affect existing installations and compatibility.
Consider one of these approaches:
- Add a prominent note in the Getting Started section about the version change
- Test compatibility with Python 3.8-3.9 to see if you can maintain backward compatibility
- Document migration steps for users on older Python versions
Add interactive mode for continuous task execution
There was a problem hiding this comment.
Actionable comments posted: 2
🔭 Outside diff range comments (1)
README.md (1)
7-9:⚠️ Potential issuePython version requirements are inconsistent
The badge at the top advertises Python 3.8+, but the “Prerequisites” section
later says Python 3.10+. Please align these two statements (and any
pyproject.toml/setup.cfgclassifiers) to avoid user confusion and
installation failures.Example fix:
-[](https://www.python.org) +[](https://www.python.org) @@ -**Python 3.10+** +**Python 3.10+ (required)**Also applies to: 147-151
🧰 Tools
🪛 markdownlint-cli2 (0.17. F438 2)
9-9: No empty links
null(MD042, no-empty-links)
♻️ Duplicate comments (1)
README.md (1)
158-159: Spelling: use “Node.js”The correct capitalization is “Node.js” (per LanguageTool hint).
-**Node.js & npm/npx:** +**Node.js & npm/npx:**🧰 Tools
🪛 LanguageTool
[uncategorized] ~158-~158: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...(NODE_JS)
🧹 Nitpick comments (4)
ra_aid/__main__.py (3)
1151-1171: Guard against missingexamples/default_mcp_servers.jsonin source distributions
importlib.resources.files("ra_aid").joinpath("examples/default_mcp_servers.json")
assumes the file is shipped as package data. If the user installs RA.Aid from a
wheel withoutpackage_data,is_file()returnsFalseand you log an error,
but later logic still treatsdefault_config_dictas valid (empty dict) and
proceeds to enable MCP-Use. This silently disables Context7/TaskMaster without
informing the user why.Consider:
- Explicitly setting
mcp_enabled = Falsewhen defaults are missing and
--mcp-use-configwasn’t provided.- Emitting a one-shot warning explaining that default MCP servers were skipped
because the JSON file is absent.This improves transparency and avoids user confusion.
958-982: Interactive REPL swallowsKeyboardInterruptinside inner loop
run_interactive_mode()has a nestedtry/except (EOFError, KeyboardInterrupt)
inside thewhile True, then an outertry/except Exception. As a result,
Ctrl-Conly exits the current task but drops the user back to the prompt,
contradicting the help text (“Press Ctrl+C to exit.”).If the intent is to terminate RA.Aid on
Ctrl-C, break the loop or re-raise:except (EOFError, KeyboardInterrupt): - print("\n👋 Goodbye!") - sys.exit(0) + print("\n👋 Goodbye!") + break # or: raiseOtherwise, update the welcome text to clarify that a double
Ctrl-C(orCtrl-D)
is required.
1231-1248: Simplify nestedifstatements (ruff SIM102)The consecutive checks for missing environment variables can be flattened:
-if task_master_planning_enabled: - if not os.getenv("ANTHROPIC_API_KEY"): - logger.warning("…") +if task_master_planning_enabled and not os.getenv("ANTHROPIC_API_KEY"): + logger.warning("…")The same pattern applies to the GitHub token block below.
🧰 Tools
🪛 Ruff (0.8.2)
1239-1240: Use a single
ifstatement instead of nestedifstatements(SIM102)
1245-1246: Use a single
ifstatement instead of nestedifstatements(SIM102)
README.md (1)
158-163: Indentation breaks Markdown list renderingThe four OS-specific commands under “C/C++ Build Tools” are indented with 4
spaces instead of 2, trippingmarkdownlint(MD007) and rendering as a single
code block rather than list items on GitHub.- - **Debian/Ubuntu:** `sudo apt-get update && sudo apt-get install build-essential` - - **Fedora:** `sudo dnf groupinstall "Development Tools"` - - **macOS:** Install Xcode Command Line Tools: `xcode-select --install` - - **Windows:** Install Visual Studio with C++ build tools (select "Desktop development with C++" workload in the installer). + - **Debian/Ubuntu:** `sudo apt-get update && sudo apt-get install build-essential` + - **Fedora:** `sudo dnf groupinstall "Development Tools"` + - **macOS:** Install Xcode Command Line Tools: `xcode-select --install` + - **Windows:** Install Visual Studio with C++ build tools (select "Desktop development with C++" workload in the installer).🧰 Tools
🪛 LanguageTool
[uncategorized] ~158-~158: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...(NODE_JS)
🪛 markdownlint-cli2 (0.17.2)
160-160: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
161-161: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
162-162: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
163-163: Unordered list indentation
Expected: 2; Actual: 4(MD007, ul-indent)
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
README.md(6 hunks)ra_aid/__main__.py(13 hunks)
🧰 Additional context used
🪛 Ruff (0.8.2)
ra_aid/__main__.py
1216-1216: f-string without any placeholders
Remove extraneous f prefix
(F541)
1239-1240: Use a single if statement instead of nested if statements
(SIM102)
1245-1246: Use a single if statement instead of nested if statements
(SIM102)
🪛 LanguageTool
README.md
[uncategorized] ~158-~158: The official spelling of this programming framework is “Node.js”.
Context: ... (Context7, Task Master). Install from nodejs.org. - **C/C++ Bu...
(NODE_JS)
[uncategorized] ~247-~247: Loose punctuation mark.
Context: ... Command Line Options - -m, --message: The task or query to be executed (optio...
(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~259-~259: Loose punctuation mark.
Context: .... - --disable-default-mcp [SERVER ...]: Disable specific default MCP servers (c...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~259-~259: To form a complete sentence, be sure to include a subject.
Context: ..., github) or all if no names are given. Can be used with --mcp-use-config to merg...
(MISSING_IT_THERE)
[uncategorized] ~260-~260: Loose punctuation mark.
Context: ...ults. - --disable-task-master-planning: Force RA.Aid to use its internal planne...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~317-~317: Consider an alternative adjective to strengthen your wording.
Context: ... Abstract Syntax Trees (ASTs), enabling deeper understanding of code relationships, sy...
(DEEP_PROFOUND)
[uncategorized] ~323-~323: Loose punctuation mark.
Context: ...able-default-mcp context7 taskmaster-ai: Disable specific default servers. - --...
(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~324-~324: Loose punctuation mark.
Context: ...s. - --mcp-use-config ./my_config.json: Use only a custom MCP configuration. - ...
(UNLIKELY_OPENING_PUNCTUATION)
[uncategorized] ~325-~325: Loose punctuation mark.
Context: ...tion. - --disable-task-master-planning: Keep Task Master tools available but us...
(UNLIKELY_OPENING_PUNCTUATION)
[style] ~327-~327: This wording can make your sentence hard to follow. Try rephrasing for improved clarity.
Context: ...ult MCP server fails to initialize (e.g., due to missing dependencies), it will be autom...
(DUE_TO_BECAUSE)
🪛 markdownlint-cli2 (0.17.2)
README.md
160-160: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
161-161: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
162-162: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
163-163: Unordered list indentation
Expected: 2; Actual: 4
(MD007, ul-indent)
ra_aid/__main__.py
Outdated
| def process_task(args): | ||
| """Process a single task with the given arguments.""" | ||
| # This function will contain the core task processing logic | ||
| # that was previously in main() | ||
|
|
||
| # Create config repository for storing runtime configuration | ||
| config_repo = get_config_repository() | ||
|
|
||
| # Store key arguments in config repository for access by tools | ||
| config_repo.set("provider", args.provider) | ||
| config_repo.set("model", args.model) | ||
| config_repo.set("temperature", args.temperature) | ||
| config_repo.set("expert_provider", args.expert_provider) | ||
| config_repo.set("expert_model", args.expert_model) | ||
| config_repo.set("research_only", args.research_only) | ||
|
|
||
| # Safely set web_research and search flags | ||
| web_research_enabled = getattr(args, 'web_research', False) | ||
| search_enabled = getattr(args, 'search', False) | ||
| config_repo.set("web_research_enabled", web_research_enabled) | ||
| config_repo.set("search_enabled", search_enabled) | ||
|
|
||
| config_repo.set("hil", args.hil) | ||
| config_repo.set("test_cmd_timeout", args.test_cmd_timeout) | ||
| config_repo.set("max_test_cmd_retries", args.max_test_cmd_retries) | ||
|
|
||
| # Initialize MCP-Use integration if enabled | ||
| mcp_use_client_instance = None | ||
| if hasattr(args, 'mcp_use_config') and args.mcp_use_config: | ||
| try: | ||
| from ra_aid.utils.mcp_use_client import MCPUseClientSync | ||
| logger.info("Attempting to initialize MCP-Use client...") | ||
| mcp_use_client_instance = MCPUseClientSync(args.mcp_use_config) | ||
| active_mcp_servers = mcp_use_client_instance.get_active_server_names() | ||
| logger.info(f"MCP-Use client initialized. Active servers: {active_mcp_servers}") | ||
| config_repo.set("mcp_use_enabled", True) | ||
| config_repo.set("active_mcp_servers", active_mcp_servers) | ||
| except Exception as e: | ||
| logger.error(f"MCP-Use client initialization failed: {e}") | ||
| logger.warning("MCP-Use integration will be disabled for this run.") | ||
| config_repo.set("mcp_use_enabled", False) | ||
| config_repo.set("active_mcp_servers", []) | ||
|
|
There was a problem hiding this comment.
🛠️ Refactor suggestion
process_task() is executed without an active DB/repository context
process_task() calls repository getters (get_config_repository(), get_session_repository(), get_human_input_repository(), …).
When invoked from interactive mode (run_interactive_mode()), it is outside the with DatabaseManager(...): block created in main(), meaning all repository managers have already been closed. This will trigger database-closed exceptions on the second interactive request.
A minimal fix is to wrap process_task() itself in the same context managers you use in main():
def process_task(args):
- # existing code...
+ from ra_aid.database import DatabaseManager
+ from ra_aid.database.repositories.session_repository import (
+ SessionRepositoryManager,
+ )
+ # … import the other *RepositoryManager classes you need …
+
+ with DatabaseManager(base_dir=args.project_state_dir) as db, \
+ SessionRepositoryManager(db), \
+ KeyFactRepositoryManager(db), \
+ KeySnippetRepositoryManager(db), \
+ HumanInputRepositoryManager(db), \
+ ResearchNoteRepositoryManager(db), \
+ TrajectoryRepositoryManager(db), \
+ WorkLogRepositoryManager(), \
+ ConfigRepositoryManager():
+ # ── all existing process_task logic stays here ──Alternatively, move the interactive loop entirely inside the original with block and keep repository managers alive for the whole REPL session.
Failing to address this will make the new interactive mode unstable after the first task.
Committable suggestion skipped: line range outside the PR's diff.
ra_aid/__main__.py
Outdated
| # Eagerly validate file paths provided via arguments | ||
| if parsed_args.msg_file and not os.path.exists(parsed_args.msg_file): | ||
| parser.error(f"Message file not found: {parsed_args.msg_file}") | ||
| if parsed_args.custom_tools and not os.path.exists(parsed_args.custom_tools): | ||
| parser.error(f"Custom tools file not found: {parsed_args.custom_tools}") | ||
| if parsed_args.mcp_use_config and not os.path.exists(parsed_args.mcp_use_config): | ||
| parser.error(f"MCP-Use config file not found: {parsed_args.mcp_use_config}") | ||
| if parsed_args.aider_config and not os.path.exists(parsed_args.aider_config): | ||
| parser.error(f"Aider config file not found: {parsed_args.aider_config}") | ||
|
|
There was a problem hiding this comment.
Use os.path.isfile() for strict path validation
os.path.exists() also returns True for directories.
Because the CLI flags (--msg-file, --custom-tools, --mcp-use-config, --aider-config) are all file arguments, accidentally passing a directory will skip validation and fail later with less helpful errors.
-if parsed_args.msg_file and not os.path.exists(parsed_args.msg_file):
+if parsed_args.msg_file and not os.path.isfile(parsed_args.msg_file):
@@
-if parsed_args.custom_tools and not os.path.exists(parsed_args.custom_tools):
+if parsed_args.custom_tools and not os.path.isfile(parsed_args.custom_tools):
@@
-if parsed_args.mcp_use_config and not os.path.exists(parsed_args.mcp_use_config):
+if parsed_args.mcp_use_config and not os.path.isfile(parsed_args.mcp_use_config):
@@
-if parsed_args.aider_config and not os.path.exists(parsed_args.aider_config):
+if parsed_args.aider_config and not os.path.isfile(parsed_args.aider_config):This change prevents silent acceptance of invalid directory paths and surfaces a clear, early CLI error.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| # Eagerly validate file paths provided via arguments | |
| if parsed_args.msg_file and not os.path.exists(parsed_args.msg_file): | |
| parser.error(f"Message file not found: {parsed_args.msg_file}") | |
| if parsed_args.custom_tools and not os.path.exists(parsed_args.custom_tools): | |
| parser.error(f"Custom tools file not found: {parsed_args.custom_tools}") | |
| if parsed_args.mcp_use_config and not os.path.exists(parsed_args.mcp_use_config): | |
| parser.error(f"MCP-Use config file not found: {parsed_args.mcp_use_config}") | |
| if parsed_args.aider_config and not os.path.exists(parsed_args.aider_config): | |
| parser.error(f"Aider config file not found: {parsed_args.aider_config}") | |
| # Eagerly validate file paths provided via arguments | |
| if parsed_args.msg_file and not os.path.isfile(parsed_args.msg_file): | |
| parser.error(f"Message file not found: {parsed_args.msg_file}") | |
| if parsed_args.custom_tools and not os.path.isfile(parsed_args.custom_tools): | |
| parser.error(f"Custom tools file not found: {parsed_args.custom_tools}") | |
| if parsed_args.mcp_use_config and not os.path.isfile(parsed_args.mcp_use_config): | |
| parser.error(f"MCP-Use config file not found: {parsed_args.mcp_use_config}") | |
| if parsed_args.aider_config and not os.path.isfile(parsed_args.aider_config): | |
| parser.error(f"Aider config file not found: {parsed_args.aider_config}") |
Fix interactive mode issues and add MCP servers display
Fix interactive mode issues (v2)
Ensures MCP servers are correctly initialized and available in interactive mode.
When user selects 'n' for a shell command, prompt for feedback.
Feat/interactive mode fixes
fix: Display status panel immediately in interactive mode
fix: Ensure MCP status displays correctly in interactive mode
chore: Add debug logging for MCP server status
Ensures MCP servers are initialized before interactive mode decision.
refactor: Initialize MCP servers before interactive mode check
feat: Add wait time and message for MCP server initialization
chore: Add detailed logging to MCP config loading
fix: Correct path for loading default MCP config file
| Using MCP servers often requires external dependencies managed by that | ||
| server. For example, using the Context7 server requires ``npx`` (Node.js) | ||
| to be available in the environment to run ``@upstash/context7-mcp``. | ||
|
|
There was a problem hiding this comment.
looks like this docstring got mangled a bit. LangChainAdapter got cut off from the preceding paragraph.
|
@noodleA1 assume I have RA.Aid set up and running, e.g. I have What's the best way to test this? |
|
Awesome PR btw --we'll have to get it reviewed but improving MCP support is a great contribution 👍 |
| If it does, use `task-master list` and `task-master show` to understand the existing plan and align with it. | ||
| If the user provided a PRD file path, prioritize using `task-master parse-prd <filepath>` to generate the initial task list. | ||
| When generating or refining complex steps, consider using `task-master expand --id=<task_id>`. | ||
| After a task implementation is successfully completed (i.e., `request_task_implementation` returns `success: true` and a `task_master_id`), you MUST immediately call `task-master set-status --id=<returned_task_master_id> --status=done` to keep the Task Master state synchronized. |
There was a problem hiding this comment.
Question: If we always need to set status to done on success and we have the id of that task - why don't we automatically call set-status ourselves internally instead of relying on the AI to do it?
| @@ -0,0 +1,7 @@ | |||
|
|
|||
| Tree-sitter Tool Guidance: | |||
| Use tree-sitter tools (`get_ast`, `get_symbols`, `run_query`, `get_dependencies`) for structural code analysis. These complement text search (`run_shell_command` with `rg`). For example: | |||
There was a problem hiding this comment.
Shouldn't we expand on this to explain more details like params or give examples?
get_ast(project="my-project", path="src/main.py", max_depth=5, include_text=True)
Unless directions are given somewhere else as well?
| "reason": reason, | ||
| "agent_crashed": agent_crashed, | ||
| "crash_message": crash_message, | ||
| "task_master_id": task_master_id # Pass back the ID | ||
| } |
There was a problem hiding this comment.
So what happens if we don't return back key facts or snippets? Are we not tracking those anymore with this new change for taskmaster implementation?
|
@noodleA1 Have you got this fully integrated and running yet? If so what are your general thoughts of it? Is it running better or worse with these MCP tools? |
|
Still working on it, but took a break from it for something else. I'll likely lean back into it, but was testing plandex as it recently became active again. I had some pretty big issues with this PR as it stands but looking forward to adjusting it. So far EVERY ai system with sequential thinking, context7, and even firecrawl/brave/sometimes taskmaster has improved dramatically, or at least confidence has risen which is no small thing with these systems |
Adjusted the mcp tool to use the repo mcp-use from pietrozullo and the mcp server context7 from upstash to make a more extensible and compatible mcp capable agent, giving the agent the capacity to pull up to date documentation from thousands of packages using context 7
Summary by CodeRabbit
New Features
Enhancements
Bug Fixes
Chores