Skip to content
| Marketplace
Sign in
Visual Studio Code>Programming Languages>ECIP Local — Code IntelligenceNew to Visual Studio Code? Get it now.
ECIP Local — Code Intelligence

ECIP Local — Code Intelligence

Preview

Naman09Khater

|
5 installs
| (0) | Free
Local-first code intelligence for large repositories. Indexes your workspace and exposes an MCP server for AI agents (GitHub Copilot, Cursor, Claude Code).
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

ECIP Local — Code Intelligence

Local-first code intelligence for large repositories. Indexes your workspace with Tree-sitter, stores everything in SQLite, and exposes an MCP server so GitHub Copilot, Cursor, Claude Code, and other AI agents can query your code — no cloud, no Docker, no API keys.


Features

  • Symbol search — find any function, class, interface, or type by name across your workspace
  • Symbol resolution — get full details: docstring, type signature, file, line, reference count
  • Reference lookup — find everywhere a symbol is called or used
  • Semantic search — natural-language queries over code chunks (ONNX embeddings or BM25 fallback)
  • Dependency inspector — list manifest dependencies (npm, Cargo, go.mod, Maven, pip)
  • File context — get all chunks and symbols for a specific file
  • Incremental indexing — file saves trigger instant re-indexing of only the changed file
  • Zero-config Copilot integration — auto-writes .vscode/mcp.json for GitHub Copilot

Supported Languages

TypeScript · JavaScript · Python · Rust · Go · Java

Quick Start

  1. Install the extension from the VS Code Marketplace (or sideload the .vsix)
  2. Open any workspace containing supported source files
  3. The extension activates automatically and begins indexing in the background
  4. The status bar shows live progress: ⟳ ECIP: 42/170 (24%)
  5. When complete: ✓ ECIP: 170 files · 3,412 symbols
  6. GitHub Copilot Chat can now use the 6 MCP tools — try asking about your code

MCP Tools

Tool Description
search_symbols Pattern-match symbol FQNs (e.g. auth, *Token*)
resolve_symbol Get docstring, type sig, file, line, ref count for an exact FQN
find_references Find all call sites of a symbol
semantic_search NLP search over code chunks
get_dependencies List manifest dependencies by ecosystem
get_file_context Get chunks + symbols for a file path

Commands

Command Description
ECIP: Reindex Workspace Force a full re-index
ECIP: Show Index Status Show MCP server port and status
ECIP: Configure MCP Rewrite .vscode/mcp.json
ECIP: Clear Index Delete the local index (re-indexed on next open)

Settings

Setting Default Description
ecip-local.embedding.mode auto auto tries ONNX then BM25; onnx requires ONNX; fallback forces BM25
ecip-local.indexing.fileExtensions .ts .tsx .js ... File extensions to index (restart required)
ecip-local.indexing.excludePatterns **/node_modules/** ... Glob patterns to skip
ecip-local.indexing.maxFileSizeKb 512 Skip files larger than this (KB)
ecip-local.indexing.maxConcurrency 4 Concurrent files during indexing
ecip-local.mcp.port 0 MCP server port (0 = auto-assign)
ecip-local.mcp.autoConfigureCopilot true Auto-write .vscode/mcp.json

How It Works

Workspace files
      │
      ▼
  Tree-sitter WASM (TypeScript / JS / Python / Rust / Go / Java)
      │
      ▼
  Symbol extraction + chunking (ecip-core)
      │
      ▼
  Embeddings (ONNX all-MiniLM-L6-v2 · 384 dims, or BM25 fallback)
      │
      ▼
  SQLite (WAL mode, FTS5, optional sqlite-vec ANN)
      │
      ▼
  MCP HTTP+SSE server  ←→  GitHub Copilot / Cursor / Claude Code

The index is stored in VS Code's extension global storage — one .db file per workspace, never committed to your repo.

Performance

Tested on a ~170-file React/TypeScript workspace:

  • Initial index: < 60 seconds
  • Incremental re-index on file save: < 500 ms
  • Symbol search: < 10 ms
  • Semantic search (BM25): < 50 ms

Troubleshooting

Extension doesn't activate — make sure the workspace contains .ts, .js, .py, .rs, .go, or .java files, or a package.json / Cargo.toml / go.mod.

semantic_search uses BM25 instead of ONNX — first run downloads the all-MiniLM-L6-v2 model (~23 MB) to extension storage. Watch the status bar for ⬇ ECIP: Downloading model X%. If download fails, the extension falls back to BM25 permanently until you retry.

No results from MCP tools — check the ECIP Local output channel for errors. Common causes: indexing still in progress, workspace not yet scanned, or large files exceeding maxFileSizeKb.

Native module error in output channel — run npm run rebuild in the extension directory to recompile better-sqlite3 for VS Code's Electron version. The extension falls back to sql.js automatically if rebuild hasn't been run, but vector search is disabled in that mode.

License

MIT

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft