code-indexer MCP server
Commit Nov 10, 2026Enables AI clients to perform semantic, full-text, and structural code searches across repositories using vector embeddings and SCIP indexes.
Also known as LightspeedDMS/code-indexer
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 10, 2026
- License
- MIT
What is code-indexer MCP?
code-indexer enables AI clients to perform semantic, full-text, and structural code searches across repositories using vector embeddings and SCIP indexes.
Summary generated from the project's README
About code-indexer MCP server
Code Indexer, also known as cidx, provides developers and AI coding assistants with advanced search capabilities for git repositories. It creates local indexes within a project directory using Tantivy for full-text searches and HNSW for vector embeddings. Users can perform semantic, keyword, regex, and hybrid queries to find code based on meaning rather than just specific text strings. The system supports Python 3.9 through 3.12 and requires a VoyageAI API key or Cohere embeddings to function.
This tool serves individual developers via a command line interface and teams through a centralized server deployment. The server version offers a Web UI, REST API, and multi-user support with SSO and MFA. Beyond simple text matching, it handles structural AST searches using X-Ray evaluators and code intelligence through SCIP indexes. It can also index git history for temporal searches and map dependencies across multiple repositories.
What you can do with it
- 1Semantic code discoveryAsk an AI to find specific logic within a codebase using natural language descriptions of functionality.
- 2Structural code analysisPerform deep code intelligence tasks like finding definitions, references, and call chains.
- 3Historical code searchingLocate changes or specific code states within a certain timeframe using git history.
- 4Hybrid search queriesCombine exact keyword matching with semantic meaning to refine search results.
Written from the project's README. Every use case is checked against a quote from it.
Configuration
VOYAGE_API_KEY
| Name | Description | |
|---|---|---|
| VOYAGE_API_KEY | — |
README
View on GitHubCode Indexer (cidx)
Semantic code search for your repositories: find code by what it does, not only by the words it contains.
What it is
CIDX indexes a git repository into a local vector index (VoyageAI or Cohere embeddings, HNSW) and a full-text
index (Tantivy), and answers natural-language, keyword and regex queries over it. It can also index git history,
build SCIP code-intelligence indexes, and run structural AST searches. Everything lives in the project's
.code-indexer/ directory; nothing runs in a container.
It is for:
- developers who want to search their own code by meaning from the command line;
- AI coding assistants, through instructions installed by
cidx teach-aior through the server's MCP endpoint; - teams, through the CIDX server: shared, centrally indexed repositories with REST, MCP and a Web UI.
Quickstart
Requires Python 3.9-3.12, a C/C++ compiler (for the hnswlib build) and a VoyageAI API key. Details: Installation.
pipx install git+https://github.com/LightspeedDMS/code-indexer.git@master
export VOYAGE_API_KEY="your-voyage-key"
cd /path/to/your/git/repository
cidx init
cidx index --fts
cidx query "where are requests rejected when the limit is exceeded" --limit 5
cidx query "authenticate_user" --fts # exact text
cidx query "token bucket" --fts --semantic # hybrid: keyword and meaning
Each semantic result shows the score, file and line range, then the matching code; add --quiet for a compact
listing.
Capabilities
- Semantic, full-text, regex and hybrid search with language, path and score filters: Query guide
- Git history search with time ranges: Temporal search
- Definitions, references, call chains and impact analysis (SCIP): SCIP guide
- Structural AST search with user-written evaluators (X-Ray): X-Ray cookbook
- A background daemon with in-memory indexes and watch mode: Operating modes
- Instructions and skills for local AI assistants: teach-ai
- Embedding providers (VoyageAI, Cohere) and project settings: Configuration
- Multi-user server with golden repositories, REST API, Web UI, SSO and MFA: Server deployment
- MCP endpoint for AI assistants: MCP registration
- Cross-repository dependency map: Meta-repo discovery
- Searchable Langfuse traces: Langfuse trace sync
- Multi-node cluster on PostgreSQL: Cluster setup
All documentation, by audience: docs/README.md. Design: Architecture overview.
Project
- Contributing: CONTRIBUTING.md and the Code of Conduct
- Security reports: privately, as described in SECURITY.md; not in public issues
- Bugs and feature requests: GitHub Issues
- Questions: GitHub Discussions
- Release history: CHANGELOG.md
- License: MIT
code-indexer MCP questions
What kind of searches can I perform?
You can perform semantic, full-text, regex, and hybrid searches, as well as git history searches and structural AST searches.
Does it require Docker or containers?
No, everything is stored in the project's local directory and nothing runs in a container.
Which embedding providers are supported?
The tool supports VoyageAI and Cohere for embedding generation.
Can I use this with Claude Desktop?
Yes, it provides an MCP endpoint specifically for AI assistants like Claude Desktop.