Kodit MCP
A server that indexes Git repositories and documents and gives an assistant search over MCP: BM25, semantic, grep, visual
- 127 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The server clones and indexes Git repositories and search is open without a key by default.
- Enrichments pass source code to the configured LLM provider.
Install
- Works in
Manual install
docker run -p 8080:8080 registry.helix.ml/helix/kodit:latestThe first README option: SQLite and the built-in embedding model, no API keys needed.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server clones and indexes Git repositories and search is open without a key by default.
- Enrichments pass source code to the configured LLM provider.
This is third-party code. Review the repository files before installing.
What it does
Kodit splits Git repositories into snippets and serves them to MCP-compatible assistants. Before writing code, the assistant asks Kodit for examples from your own projects and gets up-to-date snippets. Separate tools cover different tasks: `kodit_keyword_search` (BM25), `kodit_semantic_search`, `kodit_grep`, `kodit_ls`, `kodit_read_resource` and visual search with `kodit_visual_search`. Documents are indexed too: PDF, Word, PowerPoint and spreadsheets are rasterised, so search covers code and documentation together. Optional LLM enrichments produce architecture, API and database schema docs, usage examples and commit descriptions. Basic operation needs no external services: an embedding model and SQLite are built in.
Who it is for. Teams whose code and documentation live in several repositories and whose assistant should write code from examples in their own projects.
Good fit when
- The assistant invents an API from scratch though a sibling company repository already has a working example.
- A snippet has to be found by meaning when keywords do not locate it.
- Documentation lives in PDFs and slide decks and the answer must point to a page.
- A new developer needs an architecture map of a repository and a database schema.
Not a fit when
- There is a single small repository: the assistant's built-in search is enough.
- Source code cannot go to an LLM for enrichments: those tools work only with a configured provider.
- Search endpoints need protection: they are open by default.
Example request
Index the repository, find examples of order handling in our projects through Kodit and write a new handler following them.MCP
- Transport
- http
- Authentication
- not required
Environment variables
API_KEYSsecret- Comma-separated API keys for write operations (creating repositories, triggering syncs). Search is open by default.
Limitations
Enrichment tools (`kodit_architecture_docs`, `kodit_api_docs`, `kodit_database_schema`, `kodit_cookbook`, `kodit_wiki` and `kodit_commit_description`) work only with a configured LLM provider. Keys in `API_KEYS` guard write operations while search is open by default, so an exposed server should be closed at the network level. For production the authors recommend PostgreSQL with VectorChord instead of SQLite. Some assistants do not call Kodit on their own, and a rule in the project instructions is then needed.
How to disable. Remove the `kodit` entry from the client's MCP settings and stop the container or the `kodit serve` process. Index data sits in `~/.kodit` and is deleted separately.
FAQ
How do I connect Kodit to Claude Code?
Run `claude mcp add --transport http kodit http://localhost:8080/mcp`. If the server runs elsewhere, replace the address with your own.
Are API keys needed?
Not for basic operation: the Docker image starts with SQLite and a built-in embedding model. An LLM key is needed only for enrichments, and `API_KEYS` guard write operations.
Which clients are supported?
The README has setups for Claude Code, Cursor, Cline and Kilo Code. Any client with MCP over HTTP will work.
What if the assistant does not call Kodit?
Add a rule to project rules or the system prompt requiring Kodit search tools before writing code. In Cursor that is `.cursor/rules/kodit.mdc` with `alwaysApply: true`.
Can Kodit be used as a Go library?
Yes. The `github.com/helixml/kodit` package embeds into a Go application, which is how it is integrated into the Helix platform. A generated HTTP client exists for a remote server.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server that indexes a local repository with tree-sitter and gives the agent code search, file lookup and a structure summary of a file
An MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
A local codebase index for agents: hybrid search, dependency graphs and change impact analysis
codesearch searches code across several repositories with hybrid retrieval and runs fully offline, without a GPU or Docker.
Docs MCP Server
Grounded Docs MCP Server
A local MCP server that indexes docs from websites, GitHub, npm, PyPI and files and gives the agent version-aware search
A local semantic code index: the agent searches by meaning, then follows hits into definitions, callers and dependency paths
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Kodit (this card) | no | not required | 127 | |
| Claude Context | yes | API key | 12,690 | claude-context builds a vector index of one codebase; Kodit gathers several repositories and documents and adds grep, BM25 and visual search. |
| code-index-mcp | no | not required | 162 | code-index-mcp stores a code index in SQLite with a call graph; Kodit targets examples from your projects and PDF and Office documents. |
README in short
The README explains the server's purpose, a quick start with Docker or a binary, connection to four assistants, a list of 15 MCP tools, the REST API, use as a Go library and production setup.