Lightweight MCP server for AI IDEs (Claude Code, Cursor, etc.). Provides 21 tools across 4 modules without LLM wrapping.
pip install -r requirements.txt
python server.pyindex_project- Index project files and extract symbolsget_symbol_info- Look up symbol informationget_dependency_graph- Analyze file dependenciesget_project_stats- Get project statistics
search_docs- Search documentation (stub)get_package_info- Query package registries (PyPI/npm)search_code_examples- Search code examples (stub)check_compatibility- Check dependency compatibility
git_status- Get git working tree statusgit_history- Get commit historygit_branch_analysis- Analyze branches (stub)ci_status- Check CI status (stub)issue_list- List GitHub issues (stub)pr_summary- Get PR summary (stub)
list_specs- List specification filesget_spec- Read spec contentsearch_specs- Search specscreate_spec- Create new spec from templatescaffold_project- Scaffold project from templatevalidate_structure- Validate directory structure (stub)
- CLI-first: All external calls via
git,gh,pip,npm,curl - Zero LLM wrapping: Return structured data, let AI IDE analyze
- File-based cache: TTL cache in
~/.cache/ai-coding-mcp/ - Pydantic config: Type-safe configuration
ai-coding-mcp/
├── server.py # FastMCP entry, tool registration
├── config.py # Pydantic config
├── tools/
│ ├── context/ # code_indexer, dependency_graph, project_stats
│ ├── knowledge/ # doc_search, package_info, code_search, compatibility
│ ├── workflow/ # git_ops, ci_github
│ └── specs/ # spec_manager, scaffold, validator
├── utils/
│ ├── cli_runner.py # Unified CLI executor
│ └── cache.py # File-based TTL cache
├── tests/
├── requirements.txt
└── README.md
pytest tests/ -vFor Claude Code, add to your MCP settings:
{
"mcpServers": {
"ai-coding": {
"command": "python",
"args": ["server.py"]
}
}
}MIT License

