Skip to main content

Model Context Protocol server for IDE-like code navigation and semantic search

Project description

Sigil MCP Server Version Tests Coverage Changelog

A Model Context Protocol (MCP) server that provides IDE-like code navigation and search for local repositories. Gives AI assistants like ChatGPT powerful code exploration capabilities including symbol search, trigram indexing, and semantic navigation.

Quickstart

See docs/QUICKSTART.md for the fastest path to a working config and the most common knobs (index path, repos, embeddings on/off, admin settings).

Features

Hybrid Code Search

  • Fast text search using trigram indexing (inspired by GitHub's Blackbird)
  • Trigram store uses RocksDB via rocksdict (install with pip install -e .[trigrams-rocksdict]); SQLite fallback is removed.
  • Symbol-based search for functions, classes, methods, and variables
  • Semantic code search with vector embeddings backed by LanceDB (ANN queries, per-repo vector stores)
  • File structure view showing code outlines
  • Automatic index updates with file watching (optional)

Production Ready

  • Thread-safe concurrent access (SQLite WAL mode + RLock serialization)
  • File watcher, HTTP handlers, and vector indexing run safely in parallel
  • No "database is locked" errors from concurrent operations
  • Admin API for operational management (index rebuilds, stats, logs)
  • Comprehensive request/response logging with header redaction

Enterprise Security

  • OAuth 2.0 authentication with PKCE support for remote access
  • Local connection bypass (no auth needed for localhost)
  • API key fallback and IP whitelisting

Available Tools

  • index_repository - Build searchable index with symbol extraction
  • search_code - Fast substring search across repositories
  • goto_definition - Find symbol definitions
  • list_symbols - View file/repo structure
  • list_mcp_tools, external_mcp_prompt - Discover external MCP tools registered into Sigil
  • build_vector_index - Generate semantic embeddings for code (optional)
  • semantic_search - Natural language code search using embeddings
  • list_repos, read_repo_file, list_repo_files, search_repo - Basic operations
  • get_index_stats, ping - Server info and health checks

Quick Start

Installation

Clone and install dependencies:

git clone https://github.com/Superuser666-Sigil/SigilDERG-Custom-MCP.git
cd SigilDERG-Custom-MCP
pip install -e .[server-full]

Default embedding runtime: llamacpp with Jina v2 code embeddings (768-dim) at ./models/jina/jina-embeddings-v2-base-code-Q4_K_M.gguf.

Install Universal Ctags for symbol extraction (optional but recommended):

macOS: brew install universal-ctags Ubuntu/Debian: sudo apt install universal-ctags Arch Linux: sudo pacman -S ctags

Configuration

Copy the example config and edit with your repository paths:

cp config.example.json config.json
# Edit config.json

Example configuration:

{
  "repositories": {
    "my_project": "/absolute/path/to/your/project",
    "another_repo": "/path/to/another/repo"
  }
}

Alternatively, use environment variables:

export SIGIL_REPO_MAP="my_project:/path/to/project;another:/path/to/another"

Running the Server

Recommended: Use the restart script (starts both MCP server and Admin UI):

./scripts/restart_servers.sh

This script will:

  • Stop any running server processes
  • Start the MCP Server on port 8000
  • Start the Admin UI frontend on port 5173
  • Run both processes with nohup so they persist after terminal closes

Manual start (MCP server only):

python -m sigil_mcp.server

Stop all servers:

./scripts/restart_servers.sh --stop

On first run, OAuth credentials will be generated. Save the Client ID and Client Secret for connecting from ChatGPT.

Connecting to ChatGPT

[!IMPORTANT] Using Cloudflare Tunnel? You must disable Bot Fight Mode or ChatGPT's OAuth will fail.
📖 See Cloudflare OAuth Issue & Solution for details.

  1. Expose via ngrok: ngrok http 8000 (or use Cloudflare Tunnel)
  2. In ChatGPT, add MCP connector with OAuth authentication
  3. Use the OAuth credentials from server startup
  4. Start using: "Search my code for async functions"

Important: The server is configured for ChatGPT compatibility:

  • DNS rebinding protection is disabled (ChatGPT sends ngrok Host headers)
  • MCP endpoint mounted at root / (not /mcp)
  • OAuth authentication remains active and required

See docs/CHATGPT_SETUP.md for detailed instructions.

Usage Examples

Once connected to ChatGPT as an MCP server:

You: "Index my project repository"
ChatGPT: Indexed 342 files, found 1,847 symbols in 3.2 seconds

You: "Find where the HttpClient class is defined"
ChatGPT: Found in project::src/http/client.py at line 45

You: "Search for async functions"
ChatGPT: Found 23 matches across 8 files

You: "Build vector index for semantic search"
ChatGPT: Indexed 856 chunks from 342 documents

You: "Find code that handles user authentication"
ChatGPT: Found 5 relevant code sections (semantic search):
  - auth/handlers.py:45-145 (score: 0.89)
  - middleware/auth.py:12-112 (score: 0.84)
  ...

Architecture

Indexing Process

  1. File scanning (skips build artifacts)
  2. Content storage with SHA-256 deduplication
  3. Symbol extraction via universal-ctags
  4. Trigram inverted index generation
  5. Compression using zlib

Storage

~/.sigil_index/
├── repos.db           # SQLite: repos, documents, symbols
├── trigrams.rocksdb/  # RocksDB trigram inverted index (default, via rocksdict)
├── lancedb/       # LanceDB vector store (per-repo code_vectors tables + PQ indexes)
└── blobs/         # Compressed content

Performance

  • Symbol lookup: O(log n) via SQLite indexes
  • Text search: O(k) where k = trigrams * documents per trigram
  • Typical query latency: 10-100ms

Security

Path Traversal Protection: All paths validated to prevent escaping repository roots

Authentication Layers: OAuth 2.0 (primary), Local bypass (localhost), API keys (fallback), IP whitelist (optional)

Protection: Source code requires authentication for remote access, OAuth credentials stored with 0600 permissions, tokens expire after 1 hour with refresh support, PKCE prevents authorization code interception

ChatGPT Compatibility: For ChatGPT MCP connector compatibility, DNS rebinding protection is disabled. This means:

  • [NO] Host header validation: Disabled (accepts ngrok domains)
  • [NO] Content-Type validation: Disabled (accepts application/octet-stream)
  • [YES] OAuth 2.0 authentication: Active and required
  • [YES] Bearer token validation: Active
  • [YES] Token expiration: Enforced

See docs/SECURITY.md for detailed security documentation.

Documentation

Setup Guides

Architecture Decision Records (ADRs)

Other

Contributing

Contributions welcome! Please see CONTRIBUTING.md for guidelines including:

  • Contributor License Agreement (CLA) - Required for all contributors
  • Developer Certificate of Origin (DCO) requirements
  • Code standards and testing requirements
  • Pull request process
  • Code of Conduct

Licensing

Sigil is dual-licensed:

  • Open Source: Available under AGPLv3 for open-source projects and private use where source sharing requirements are met.

  • Commercial: A commercial license is required for organizations who wish to run Sigil internally without open-sourcing their own applications or who need indemnification and support.

Contact me for commercial licensing options.

See LICENSE file for full AGPLv3 text.

Licensing FAQ

Q: Can I run this inside my company under AGPLv3?

A: Yes, as long as you're comfortable with AGPLv3 and its requirements. If you expose the server to users over a network (like running it as an internal service), AGPLv3 requires making the source code available to those users, including any modifications you've made.

Q: We have a "no AGPL" policy. Can we still use Sigil?

A: Yes, via a commercial license. Email davetmire85@gmail.com to discuss your needs.

Q: Why do I have to sign a CLA to contribute?

A: The Contributor License Agreement keeps the licensing story clean—AGPLv3 for the open-source community, commercial licenses for organizations that need them—without legal ambiguity about who owns what. Your contribution remains open-source under AGPLv3; the CLA just clarifies the rights.

Q: What's included in a commercial license?

A: Commercial licenses provide freedom to use Sigil internally without open-source requirements, ability to keep modifications proprietary, indemnification and support options, and clear legal status for enterprise compliance. Contact me for details and pricing.

Q: Can I use this for my personal projects?

A: Absolutely! AGPLv3 is perfect for personal projects, hobbyist use, and small teams. You only need a commercial license if you have organizational requirements that conflict with AGPL.

For more details on contributing, see CONTRIBUTING.md.

Acknowledgments

  • Trigram indexing inspired by GitHub's Blackbird search engine
  • Symbol extraction powered by Universal Ctags
  • Built on the Model Context Protocol (MCP) specification

Support

Issues: GitHub Issues Documentation: docs/ Security: docs/SECURITY.md

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

sigil_mcp_server-1.0.1.tar.gz (268.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

sigil_mcp_server-1.0.1-py3-none-any.whl (136.1 kB view details)

Uploaded Python 3

File details

Details for the file sigil_mcp_server-1.0.1.tar.gz.

File metadata

  • Download URL: sigil_mcp_server-1.0.1.tar.gz
  • Upload date:
  • Size: 268.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.12

File hashes

Hashes for sigil_mcp_server-1.0.1.tar.gz
Algorithm Hash digest
SHA256 48f013d04db0376dac00a0194d52f52d02190c3674e2ac955c8e71f1198e497f
MD5 86e286d574f77f91b4f09576f4c8479f
BLAKE2b-256 44868837f8fb1b6bfae03b799ae5fe650ecf51ccd8bb580602103ba6a934cea2

See more details on using hashes here.

File details

Details for the file sigil_mcp_server-1.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for sigil_mcp_server-1.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 bfa19bfc56032c15f9dfeb7f3320c0cf59157eb8daef224f06ba68279417515f
MD5 cfaad1478b5a53bd02c5e07880724c7b
BLAKE2b-256 6aa3ef522563868e0b96a944bf901042a9e589dc954b407279fe397fffb0a4f5

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page