CodeRAG
Integrates with OpenAI's embedding models to enable semantic code search, allowing for hybrid ranking that combines traditional TF-IDF keyword matching with vector-based semantic understanding.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@CodeRAGfind the implementation of the JWT token validation"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Locus → now part of repomap
Locus has merged into repomap, a map of your codebase for AI agents. It combines code graph, hybrid search (Locus's BM25, now on tree-sitter AST chunks), call paths, change impact and an interactive graph UI in one Rust binary. It runs locally, needs no API key, and is MIT licensed.
npx -y @sylphx/repomap setup
@sylphx/locus1.x and@sylphx/coderag1.x are thin aliases of@sylphx/repomap, so existing configs (npx -y @sylphx/locus --root=…,codebase_search) keep working. New development happens in SylphxAI/repomap. The rest of this README documents Locus 0.x.
Locus 0.x (archived)
The exact code chunk for the job
Locus is local code search for coding agents. It indexes the repository on the machine, ranks chunks with BM25, and returns the path, line range, symbol, score, and chunk text.
No API key. No vector database. No Docker. The default transport is stdio.
npx -y @sylphx/locus --root=/absolute/path/to/projectLocus does not search the process working directory. Pass --root, set LOCUS_ROOT, or pass root on the tool call. A tool root wins, then the launch --root, then LOCUS_ROOT. CODERAG_ROOT still works when LOCUS_ROOT is unset.
Claude Code:
claude mcp add locus -- npx -y @sylphx/locus --root=/absolute/path/to/projectCodex:
codex mcp add locus -- npx -y @sylphx/locus --root=/absolute/path/to/projectRelated MCP server: mcplens
Tools
Tool | Use it when | You get |
| You can name words that appear in the code | Ranked chunks. Filters apply before the limit. |
| You already have a file and a line | Other chunks that share its tokens |
{
"query": "authenticate request",
"limit": 5,
"file_extensions": ["ts"],
"path_filter": "src/auth",
"exclude_paths": ["src/auth/legacy"]
}login does not match authenticate. Ranking is lexical. Search for the words the code uses.
What is indexed
.ts .tsx .js .jsx .mjs .cjs .rs .md .py .go
Skipped path segments, by exact name: node_modules, dist, target, .git. This is not gitignore. Files larger than 1,048,576 bytes are skipped. New indexes are written to .locus/rust-index.json and .locus/file-hashes.json. Locus still reads .coderag/ when .locus/rust-index.json is absent. Do not commit either directory.
A line that looks like a symbol starts a chunk. Other files stay one chunk. The patterns are regular expressions, not a syntax tree. Symbol chunks.
Defaults that stay boring
limitdefaults to 10 and must be an integer from 1 to 100.include_contentdefaults to true.falsereturns locators only.There is no
auto,fast, orqualitysearch mode.Each call refreshes the local index. Unchanged files are a cache hit. That refresh mode is not a search profile you select.
HTTP exists only when
MCP_TRANSPORT=http. Agents should use stdio.
Install surfaces
Surface | Where |
Claude Code |
|
Codex |
|
Cursor |
|
VS Code |
|
Other MCP clients |
|
Published natives: @sylphx/locus-darwin-arm64, @sylphx/locus-darwin-x64, @sylphx/locus-linux-arm64-gnu, @sylphx/locus-linux-x64-gnu. Windows is not a published target.
npx -y @sylphx/locus doctor prints the server version and whether locus-cli was found.
Companions
Product | Job |
PDF answers with page-level proof | |
Image facts and pixel evidence | |
Video timelines and timestamp evidence | |
Repository architecture and impact | |
Web research with source excerpts |
Locus finds the chunk. Spine maps the architecture. Install only what the task needs.
Docs: https://sylphxai.github.io/locus/
Development
bun install
bun run build:rust
cargo test
bun run docs:buildLicense
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
An MCP server that gives your AI access to the source code and docs of all public github repos
Driflyte MCP server which lets AI assistants query topic-specific knowledge from web and GitHub.
Capability registry for the agentic economy. Semantic search over verified MCP server listings.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceA high-performance MCP server for semantic search and codebase indexing using the Qdrant vector database. It features optimized embedding pipelines, AST-aware chunking, and git metadata enrichment for fast, privacy-focused local or remote search.409 npm23MIT
- AlicenseNot gradedqualityBmaintenanceA local MCP server that provides AI coding assistants with semantic search capabilities over codebases. It indexes code using local embeddings and exposes tools for efficient code retrieval, saving tokens and improving response quality.31 npm4MIT
- AlicenseBqualityDmaintenanceA high-performance MCP server for intelligent documentation search, proactive bug detection, and semantic analysis of codebases.2515 npmMIT
- AlicenseAqualityDmaintenanceMCP server for semantic code search that indexes your codebase and allows AI editors to search using natural language queries.958 npm53MIT