Deep Code Reasoning MCP Server

Performs complementary code analysis by combining Claude Code and Google's Gemini AI.

An MCP server that pairs Claude Code with Google's Gemini AI for complementary code analysis. This server enables a multi-model workflow where Claude Code handles tight terminal integration and multi-file refactoring, while Gemini leverages its massive context window (1M tokens) and code execution capabilities for distributed system debugging and long-trace analysis.

Both Claude and Gemini can handle deep semantic reasoning and distributed system bugs. This server enables an intelligent routing strategy where:

- Claude Codeexcels at local-context operations, incremental patches, and CLI-native workflows
- Gemini 2.5 Proshines with huge-context sweeps, synthetic test execution, and analyzing failures that span logs + traces + code

The "escalation" model treats LLMs like heterogeneous microservices - route to the one that's most capable for each sub-task.

- Gemini 2.5 Pro Preview: Uses Google's latest Gemini 2.5 Pro Preview (05-06) model with 1M token context window
- Conversational Analysis: NEW! AI-to-AI dialogues between Claude and Gemini for iterative problem-solving
- Execution Flow Tracing: Understands data flow and state transformations, not just function calls
- Cross-System Impact Analysis: Models how changes propagate across service boundaries
- Performance Modeling: Identifies N+1 patterns, memory leaks, and algorithmic bottlenecks
- Hypothesis Testing: Tests theories about code behavior with evidence-based validation
- Long Context Support: Leverages Gemini 2.5 Pro Preview's 1M token context for analyzing large codebases

- Node.js 18 or later
- A Google Cloud account with Gemini API access
- Gemini API key fromGoogle AI Studio

git clone https://github.com/Haasonsaas/deep-code-reasoning-mcp.git cd deep-code-reasoning-mcp

cp .env.example .env # Edit .env and add your GEMINI_API_KEY

{ "mcpServers": { "deep-code-reasoning": { "command": "node", "args": ["/path/to/deep-code-reasoning-mcp/dist/index.js"], "env": { "GEMINI_API_KEY": "your-gemini-api-key" } } } }

- Claude Code performs initial analysisusing its strengths in multi-file refactoring and test-driven loops
- When beneficial, Claude escalates to this MCP server- particularly for:

- Analyzing gigantic log/trace dumps that exceed Claude's context
- Running iterative hypothesis testing with code execution
- Correlating failures across many microservices

Note: The tool parameters use snake_case naming convention and are validated using Zod schemas. The actual implementation provides more detailed type safety than shown in these simplified examples. Full TypeScript type definitions are available insrc/models/types.ts.

The server now includes AI-to-AI conversational tools that enable Claude and Gemini to engage in multi-turn dialogues for complex analysis:

Initiates a conversational analysis session between Claude and Gemini.

{ claude_context: { attempted_approaches: string[]; // What Claude tried partial_findings: any[]; // What Claude found stuck_description: string; // Where Claude got stuck code_scope: { files: string[]; // Files to analyze entry_points?: CodeLocation[]; // Starting points service_names?: string[]; // Services involved } }; analysis_type: 'execution_trace' | 'cross_system' | 'performance' | 'hypothesis_test'; initial_question?: string; // Optional opening question }

Continues an active conversation with Claude's response or follow-up question.

{ session_id: string; // Active session ID message: string; // Claude's message to Gemini include_code_snippets?: boolean; // Enrich with code context }

Completes the conversation and generates structured analysis results.

{ session_id: string; // Active session ID summary_format: 'detailed' | 'concise' | 'actionable'; }

Checks the status and progress of an ongoing conversation.

{ session_id: string; // Session ID to check }

Main tool for handing off complex analysis from Claude Code to Gemini.

{ claude_context: { attempted_approaches: string[]; // What Claude tried partial_findings: any[]; // What Claude found stuck_description: string; // Where Claude got stuck code_scope: { files: string[]; // Files to analyze entry_points?: CodeLocation[]; // Starting points (file, line, function_name) service_names?: string[]; // Services involved } }; analysis_type: 'execution_trace' | 'cross_system' | 'performance' | 'hypothesis_test'; depth_level: 1-5; // Analysis depth time_budget_seconds?: number; // Time limit (default: 60) }

Deep execution analysis with Gemini's semantic understanding.

{ entry_point: { file: string; line: number; function_name?: string; }; max_depth?: number; // Default: 10 include_data_flow?: boolean; // Default: true }

Analyze impacts across service boundaries.

{ change_scope: { files: string[]; service_names?: string[]; }; impact_types?: ('breaking' | 'performance' | 'behavioral')[]; }

Deep performance analysis beyond simple profiling.

{ code_path: { entry_point: { file: string; line: number; function_name?: string; }; suspected_issues?: string[]; }; profile_depth?: 1-5; // Default: 3 }

Test specific theories about code behavior.

{ hypothesis: string; code_scope: { files: string[]; entry_points?: CodeLocation[]; // Optional array of {file, line, function_name?} }; test_approach: string; }

When Claude needs deep iterative analysis with Gemini:

// 1. Start conversation const session = await start_conversation({ claude_context: { attempted_approaches: ["Checked for N+1 queries", "Profiled database calls"], partial_findings: [{ type: "performance", description: "Multiple DB queries in loop" }], stuck_description: "Can't determine if queries are optimizable", code_scope: { files: ["src/services/UserService.ts"] } }, analysis_type: "performance", initial_question: "Are these queries necessary or can they be batched?" }); // 2. Continue with follow-ups const response = await continue_conversation({ session_id: session.sessionId, message: "The queries fetch user preferences. Could we use a join instead?", include_code_snippets: true }); // 3. Finalize when ready const results = await finalize_conversation({ session_id: session.sessionId, summary_format: "actionable" });

When a failure signature spans multiple services with GB of logs:

// Claude Code: Identifies the error pattern and suspicious code sections // Escalate to Gemini when: Need to correlate 1000s of trace spans across 10+ services // Gemini: Processes the full trace timeline, identifies the exact race window

When performance degrades but the cause isn't obvious:

// Claude Code: Quick profiling, identifies hot paths // Escalate to Gemini when: Need to analyze weeks of performance metrics + code changes // Gemini: Correlates deployment timeline with perf metrics, pinpoints the exact commit

When you have theories but need extensive testing:

// Claude Code: Forms initial hypotheses based on symptoms // Escalate to Gemini when: Need to test 20+ scenarios with synthetic data // Gemini: Uses code execution API to validate each hypothesis systematically

# Run in development mode npm run dev # Run tests npm test # Lint code npm run lint # Type check npm run typecheck

┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐ │ Claude Code │────▶│ MCP Server │────▶│ Gemini API │ │ (Fast, Local, │ │ (Router & │ │ (1M Context, │ │ CLI-Native) │◀────│ Orchestrator) │◀────│ Code Exec) │ └─────────────────┘ └──────────────────┘ └─────────────────┘ │ ▼ ┌──────────────────┐ │ Code + Logs + │ │ Traces + Tests │ └──────────────────┘

- API Key: Store your Gemini API key securely in environment variables
- Code Access: The server reads local files - ensure proper file permissions
- Data Privacy: Code is sent to Google's Gemini API - review their data policies

- Ensure you've set theGEMINI_API_KEYin your.envfile or environment
- Check that the.envfile is in the project root

- Verify that file paths passed to the tools are absolute paths
- Check file permissions

- Verify your API key is valid and has appropriate permissions
- Check API quotas and rate limits
- Ensure your Google Cloud project has the Gemini API enabled

- The server uses Zod for parameter validation
- Ensure all required parameters are provided
- Check that parameter names use snake_case (e.g.,claude_context, notclaudeContext)
- Review error messages for specific validation requirements

Best Practices for Multi-Model Debugging

When debugging distributed systems with this MCP server:
- Capture the timeline first- Use OpenTelemetry/Jaeger traces with request IDs
- Start with Claude Code- Let it handle the initial investigation and quick fixes
- Escalate strategicallyto Gemini when you need:

- Analysis of traces spanning 100s of MB
- Correlation across 10+ services
- Iterative hypothesis testing with code execution

- go test -race, ThreadSanitizer for race detection
- rr or JFR for deterministic replay
- TLA+ or Alloy for formal verification
- Fork the repository
- Create a feature branch
- Make your changes
- Add tests for new functionality
- Submit a pull request

This project is licensed under the MIT License - see theLICENSEfile for details.

- Open an issue onGitHub Issues
- Check thetroubleshooting sectionabove
- Review theMCP documentation

An MCP server that gives AI agents deep understanding of C# codebases via Roslyn — type search, dependency graphs, usage analysis, and architecture overviews, all from a live in-memory compilation.