eyeye
Search

eye Docs

Public setup, usage, and integration guides for the eye MCP server.

Source browsing for coding agents

eye is a source-browsing MCP server for large local repositories. It gives coding agents a bounded project map, exact source ranges, and index-backed symbol lookup without falling back to broad file reads as the default.

Get Started Tool Reference

Quick Start

Use eye when you want an agent to:

  • map the project structure before touching code
  • read source around an exact line
  • resolve a symbol once from an anchor, then reuse symbolId
  • follow references without broad text-search fanout
  • reuse a lazy local cache instead of re-doing expensive repo scans

The shipped MCP surface stays compact:

  • get_project_structure
  • read_source_range
  • query_symbol
  • refresh_index
  • get_index_status

Core Tools

The most reliable flow for a fresh repository is:

  1. Start with get_project_structure.
  2. Read the exact file range you care about with read_source_range.
  3. Resolve a symbol with query_symbol.
  4. Reuse the returned symbolId for follow-up references or context.
  5. Refresh the local cache only when the repository changed.

Read the core tool reference

Client Setup

The server runs over stdio and fits the same local workflow across MCP-aware agents:

  • install Node.js, Corepack, Universal Ctags, and ripgrep
  • build once with pnpm install and pnpm run build
  • point your MCP client at node /absolute/path/to/eye/dist/index.js

See client integration patterns

FAQ

The current release is optimized for one project root at a time, lazy local indexing, and bounded navigation rather than whole-repository dumping.

Read the FAQ