Vexor MCP
Vexor: semantic search over files and code with an MCP server, where an agent finds a file by what it contains.
- 248 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads project files and sends their content to a remote embedding provider if one is chosen.
- Writes an index and a cache to disk.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/vexorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads project files and sends their content to a remote embedding provider if one is chosen.
- Writes an index and a cache to disk.
This is third-party code. Review the repository files before installing.
What it does
Vexor builds reusable indexes over files and code and searches by meaning when you remember what a file does but forget its name and place. The same engine is available as a Python API, a command-line tool and an MCP server. A `vexor "api client config"` search over the current directory indexes it on first run, and the result shows similarity, path, lines and a preview. The `vexor mcp` command starts a built-in MCP server over stdio with two tools. `vexor_search` searches by meaning and returns the text of found passages, so an agent rarely needs a separate file read. `vexor_index` warms the index in advance. Embedding providers are configurable: OpenAI, Gemini, VoyageAI, custom or local, with optional reranking. Local models allow offline use.
Who it is for. Developers of large repositories and note collections whose agent has to find code by meaning with no exact names or substrings for grep.
Good fit when
- The agent looks for where an API client is configured without knowing the file name.
- Before a refactor tests and helpers on the topic of config parsing are found.
- The index is warmed before work so that the first call is not slow.
- Search runs fully locally on local embedding models.
Not a fit when
- You need exact substring or regex search, which grep and ripgrep handle.
- Sending file content to an embedding provider is forbidden, and local models are needed.
- The agent already has the Vexor skill installed: take just one: the skill or MCP.
Example request
Find the files in the project where the API client is configured and show passages with config loading.MCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
VEXOR_API_KEYsecret- Embedding provider API key.
VEXOR_CONFIG_JSON- Non-secret settings as JSON, such as provider and reranking.
Limitations
It needs Python 3.11 or newer, or a ready binary from the releases. Embeddings need a provider key (`VEXOR_API_KEY`) or a local model. File content goes to the embedding provider if it is remote. The first search builds the index and takes time on a large directory, the cache is stored separately and exclusions are set in `.vexorignore`. The skill and the MCP server give the same capability, so install one per agent.
How to disable. Remove the vexor entry from the MCP configuration (`claude mcp remove vexor`) and the package with `pip uninstall vexor`. Indexes and settings live in `~/.vexor/` and are deleted by hand.
FAQ
How do I connect Vexor to an agent?
For Claude Code `claude mcp add vexor -- vexor mcp`, for Codex `codex mcp add vexor -- vexor mcp`. In other clients add the command `vexor` with the argument `mcp` to `mcpServers` and the key in `env` if needed.
How does MCP differ from the skill?
The skill teaches shell-capable agents (Claude Code, Codex) to drive the full CLI. MCP exposes search as native tools and works in any client. Choose one of the two.
Which embedding providers are supported?
OpenAI by default, Gemini, VoyageAI, custom and local models. The `--rerank` mode accepts options such as `bm25` and `hybrid`.
How do I install the skill for Claude Code or Codex?
Run `vexor install --skills claude` for Claude Code or `vexor install --skills codex` for Codex. The skill source is in `plugins/vexor/skills/vexor-cli`.
Where are settings stored?
Globally in `~/.vexor/config.json`, and the nearest `<project>/.vexor/config.json` overrides some fields. Non-secret fields can be passed through `VEXOR_CONFIG_JSON`.
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
A structural code search MCP server for agents: finds the right file by query intent and runs locally
A local code search in Rust for agents: BM25, vectors and reranking, 65 languages, offline, with a skill and MCP server.
Local search and MCP server for agents: one Rust binary that indexes code, docs and logs
codesearch searches code across several repositories with hybrid retrieval and runs fully offline, without a GPU or Docker.
A semantic code search MCP server: indexes the project into a vector database and gives the agent only relevant snippets
README in short
The `vexor` PyPI package gives a console tool, a Python API and an MCP server. Index modes (`--mode`) choose what search is built on: names, file heads, brief descriptions, full text, code or outline.