Find the exact slice
Search for behavior, not filenames. Contextro narrows the problem down to the right symbol or code chunk in milliseconds.
Single Rust binary. Local MCP server for Claude, Cursor, Windsurf, and other clients - search code by meaning, trace call graphs, inspect change impact, query git history, and keep memory across sessions without sending code to the cloud.
Install
npm install -g contextro npx contextro@latest published on npm · stdio + http · docker ready fewer tokens to find a function
fewer tokens to trace callers
cold start
MCP tools
Why it exists
Search for behavior, not filenames. Contextro narrows the problem down to the right symbol or code chunk in milliseconds.
Inspect callers, callees, impact, and commit history before renaming, deleting, or moving core code.
Return compact snippets, confidence signals, and sandboxed large outputs instead of dumping whole files into the model.
Core capabilities
Semantic search, BM25, graph signals, reranking, and AST-aware snippet compression in one local MCP toolchain.
Trace callers, callees, architecture hubs, and impact radius before you touch shared code or start a refactor.
Store decisions, recover prior sessions, and index docs or notes so agents stop restarting cold.
Install via npm, use pre-built binaries, or run in Docker. Work over stdio or HTTP without shipping code to the cloud.
What you can ask
search() Hybrid semantic + keyword + graph retrieval
impact() Inspect blast radius before edits or refactors
code() AST-based search, patterns, and rewrites
commit_search() Search git history by meaning
knowledge() Index docs and notes beside code
restore() Recover session context and project state
Works with
Claude Code
Claude Desktop
Cursor
Windsurf
Generic MCP clients
Docker / HTTP deployments
For large repositories
Hybrid retrieval and compact responses reduce how much irrelevant code an agent needs to scan before it can act.
For refactors
Call graph traversal, impact analysis, and git-aware search make shared changes safer before you touch production code.
For rollouts
Use the npm package, pre-built binaries, or the Docker image and keep the same local-first workflow across laptops and services.