bpm-code-search-mcp

Works with Claude Code, OpenCode, or any MCP client. Auto-installed by claude-experts and bpm-opencode-experts.

What it does

How it works

• Semantic search — embeds code chunks (60-line sliding window) via LM Studio (nomic-embed-text), stores in SQLite, scores with cosine similarity
• Symbol index — regex extraction at index time; no AST required; covers 10 languages
• FTS5 fallback — BM25 keyword search when no embedding provider is available
• Provider sticky — the embedding model used at index time is locked in; queries from a different provider fall back to FTS5 rather than silently mixing vector spaces

Symbol extraction covers

TypeScript/JS · Python · Go · Rust · Java · C# · Ruby · PHP · Swift · Kotlin · Markdown headings

Install

Handled automatically by claude-experts or bpm-opencode-experts install.sh.

Manual:

git clone https://github.com/bpmforge/bpm-code-search-mcp.git ~/Code/bpm-code-search-mcp
cd ~/Code/bpm-code-search-mcp && npm install && npm run build

# Claude Code
claude mcp add code-search node ~/Code/bpm-code-search-mcp/dist/index.js

# OpenCode — add to opencode.json under "mcp":
# "code-search": { "type": "local", "command": ["node", "~/Code/bpm-code-search-mcp/dist/index.js"], "enabled": true }

First use

code_index()          # index the current project (run once, then auto-updates on file edits)
code_index_status()   # verify: provider, files, chunks, symbols
code_search("authentication flow")
code_symbols(kind="class")
code_outline("src/auth.ts")
code_references("UserService")

Requirements

• Node 20–24 LTS

Embedding setup

Vector search is optional — BM25 keyword search works with no setup at all.

Default: LM Studio (free, local)

1. Download LM Studio and load nomic-ai/nomic-embed-text-v1.5-GGUF
2. No config needed — defaults point to http://localhost:1234

Alternative models

Set env vars to use a different model:

export LM_STUDIO_URL="http://localhost:1234"
export LM_STUDIO_MODEL="CompendiumLabs/bge-large-en-v1.5-gguf"  # example

OpenAI embeddings

export LM_STUDIO_URL="https://api.openai.com/v1"
export LM_STUDIO_MODEL="text-embedding-3-small"
export OPENAI_API_KEY="sk-..."

No embeddings (BM25 only)

export EMBEDDING_PROVIDER=none

Provider-sticky: The embedding model used at index time is locked in. If you change models, re-index with code_index(force=true) to rebuild from scratch.

Remote LM Studio server

export LM_STUDIO_URL="http://192.168.1.x:1234"

Environment variables

License

MIT