codecanvas-mcp 0.1.0

Precision static-analysis MCP server for Python: call graphs, control flow, and change impact

codecanvas-mcp

Precision static-analysis MCP server for Python codebases. Gives coding agents ground-truth answers about call graphs, control flow, and change impact — instead of grepping and guessing.

Tools

• list_entrypoints — API routes, scripts, and function entrypoints
• who_calls / call_tree — reverse and forward call graph, N hops
• what_does — signature, docstring, db/http/raise effects, risk
• function_flow — de-noised control-flow outline of a function
• reaching_conditions — the guard conditions behind every return/raise
• analyze_impact — changed functions and affected endpoints for a diff or git ref

Usage

claude mcp add codecanvas -- uvx codecanvas-mcp

Or in any MCP client config:

{ "mcpServers": { "codecanvas": { "command": "uvx", "args": ["codecanvas-mcp"] } } }

Pass project_path on the first tool call; it is remembered for the rest of the session.

Extras

The base install is analysis-only. pip install "codecanvas-mcp[server]" adds the FastAPI web server and runtime tracer used by the CodeCanvas VS Code extension.