A language server tells you where a symbol is defined and who references it. Asking what breaks transitively means chaining reference lookups by hand, reading files at every step. Here the dependency graph already holds the answer, so change impact is one call. The author is candid that the assistant will ignore these tools unless you instruct it in mandatory terms - soft wording gets rationalised away.
A zero-dependency indexer that parses source into structural metadata - functions, classes, imports, dependency graphs and cross-file call chains - and serves 18 query tools over it, so an agent navigates without reading whole files.
- A project summary, indexed file listing, and structural outlines for one file or the whole project
- Functions, classes and imports listed with names, lines, parameters, methods and bases
- The full source of a single function or class, fetched directly rather than by reading the file
- Symbol lookup that reports the defining file, line and kind
- Dependency answers in both directions, a transitive change-impact view, and the shortest dependency path between two symbols
- File-level import graphs both ways, and regex search across the codebase
- A forced re-index for the rare case, and session efficiency statistics
Python 3.11+ and pip install of the package with its MCP extra. PROJECT_ROOT points at the directory to index, defaulting to the working directory. In a git repository the index maintains itself: a diff check before each query re-parses only the changed files, and the index is cached to a pickle file in the project root, which you should add to .gitignore. Python is parsed with the ast module; TypeScript and JavaScript, Go, Rust and C# use regex parsing. The licence is dual - AGPL-3.0 for open-source use, commercial otherwise.
One command — pip install "mcp-codebase-index[mcp]"