vexp-mcp
v3.3.2
Published
vexp MCP server — AI context tools for coding agents
Downloads
2,505
Maintainers
Readme
vexp-mcp — MCP Server for AI Coding Agents (Claude Code, Cursor, Codex)
The Model Context Protocol server behind vexp, the local-first context engine for AI coding agents.
vexp-mcp exposes your codebase's dependency graph to any MCP-capable agent — Claude Code, Cursor, Windsurf, OpenAI Codex, Zed, Cline and more — as a small, deliberately slim set of tools. Ranked pivot files with line ranges and blast radius in one call, compact file skeletons (signatures, no bodies), and a mechanical completion check when the agent says it's done. Everything runs on your machine: no cloud indexing, no API key, and your source never leaves your laptop.
You normally don't install this package directly. Install
vexp-cli(or the VS Code extension) and runvexp setup— it wires this server into every agent it detects.
Tools your agent gets
| Tool | What it does |
|---|---|
| run_pipeline | Ranked pivot files + line ranges + blast radius for a task, in one call |
| get_skeleton | File structure (signatures, no bodies), for files the agent only needs to understand |
| verify_done | Mechanical completion check: parse errors, broken imports, untouched dependents, impacted tests |
| expand_vexp_ref | Full body of any referenced symbol, on demand |
The catalog is kept deliberately small, because every listed schema rides along on every request your agent makes.
Why teams use vexp
- Fable-5-level results from Opus 5. At half the token price. On SWE-bench Pro, Claude Code running Opus 5 with vexp resolved 81.7% (597/731); Anthropic reports 80.3% for Fable 5 and 79.2% for Opus 5. Every session published at vexp.dev/benchmark.
- One call, not a repo crawl: ranked files, line ranges and blast radius, so the agent stops rediscovering the codebase before every change.
- 100% local. Zero network calls, zero cloud indexing, works offline.
- 14 agents, 34 languages, one
vexp setup.
→ vexp.dev · Docs · vexp-cli on npm
Keywords: MCP server · Model Context Protocol · AI coding agent · Claude Code MCP · Cursor MCP · code context · token savings · dependency graph · tree-sitter · local-first · code verification
