Works with the AI Agents you already use
ContextPilot – Relevant Code Finder
Indexes JS/TS repos locally and returns only the functions relevant to your task via semantic search — cuts agent context from ~325K tokens to ~1K, no API key, no code leaves your machine.
$19
About this skill
ContextPilot automatically builds the smallest possible context for AI coding agents.
Instead of dumping an entire repository (or a keyword-matched subset of it) into an LLM's context window, ContextPilot indexes a codebase once and returns only the handful of functions actually relevant to the task at hand — as an MCP server that Claude Code, Claude Desktop, Cursor, and other MCP-compatible agents can call directly.
The problem: every AI coding agent has to figure out which part of a codebase matters for a given task. Feeding it the whole repo wastes tokens and money. Naive keyword search is cheap but shallow — it finds Login() and LoginForm() easily, but misses verifyPassword(), createJWT(), or isLoggedIn(), because those don't share the word "login" even though they're exactly where a login bug probably lives.
Measured on a real 815-file production codebase, searching for "user cannot stay logged in after refresh" (a query that never contains the word "login"): the full repo would be ~325,000 tokens of context; ContextPilot returns 5 functions, ~960 tokens — a 99.7% reduction, while still surfacing the actually relevant code.
How it works:
- Index — scans the project, parses JS/TS with the TypeScript Compiler API, stores functions/methods with exact source ranges in a local SQLite database, plus a local semantic embedding per function (all-MiniLM-L6-v2 via @xenova/transformers — no API key, nothing leaves the machine).
- Search — a natural-language task description is embedded and compared against every indexed function with cosine similarity, blended with keyword-overlap scoring.
- Call graph — caller → callee relationships are recorded during indexing; retrieval can optionally pull in direct callers/callees.
- Context — top matches are returned as pasteable, real source code, sized for the task, not the whole repo.
Included in this skill: a SKILL.md that teaches any compatible agent when to reach for ContextPilot's three MCP tools (contextpilot_index, contextpilot_search, contextpilot_context) instead of brute-force grepping or reading whole files, and how to use projectPath and call-graph expansion correctly.
Requires the contextpilotmcp MCP server (npm install -g contextpilotmcp or npx -y contextpilotmcp) to be configured in the agent's MCP settings.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
- Only JavaScript and TypeScript projects are indexed (parsed via the TypeScript Compiler API); other languages (Python, Go, Rust, Java, etc.) are not supported.
- Requires an explicit indexing step (contextpilot_index) before search/context tools return results; a freshly cloned or never-indexed project returns nothing until indexed.
- Function-level granularity only — it locates functions/methods, not config files, type-only declarations, CSS, or non-code assets.
- Semantic search quality depends on a local embedding model (all-MiniLM-L6-v2); very short, ambiguous, or highly domain-specific queries can rank unrelated functions higher than expected.
- Call-graph expansion (expandDependencies) only follows static caller/callee references the parser can resolve; dynamic dispatch, reflection, or highly dynamic code patterns may be missed.
- The index is stored locally per project (.contextpilot/index.sqlite) and is not automatically shared across machines or CI — each clone/environment must index separately.
- Re-indexing is incremental based on file mtime/size, so changes made without updating a file's modification time (rare, e.g. some sync tools) may not be picked up until a full re-index.
- Requires Node.js 22.5+ and the contextpilotmcp MCP server to be installed/configured in the client; it does not work as a passive instruction set without the MCP server running.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 10 days ago
- Software
Node.jsnpmcontextpilotmcp
Permissions
File Scopes
Reads JS/TS source files to build a function-level index (name, location, source, embedding) and a call graph. Writes the index to a local SQLite database at .contextpilot/index.sqlite inside the project root — created on first index, read on every search. No data is sent over the network at query time; embeddings run locally via @xenova/transformers.