open-code-indexer-mcp

> **Built at [«Богатый Мир Упаковки»](https://bmupack.ru), Veliky Novgorod** > MCP server for code search — ask in chat instead of opening IDE. > > **Разработано в компании [«Богатый Мир Упаковки»](https://bmupack.ru), Великий Новгород** # OpenCode Codebase Indexer Code search via BM25 + embeddings (k-NN HNSW). MCP server for AI agents. Not Google, but good enough to find your way around. ## Tools (MCP) | Tool | What it does | |------|--------------| | `search_code` | Hybrid search (BM25 + k‑NN). Set `hybrid=true` for semantic search. Filters by language, directory, package. `regex` mode. `include_source=true` to return full file body. | | `get_symbol` | Find symbol definition(s) by name. Supports multiple names (batch). `fuzzy=true` for substring matching. `include_source=true` to return symbol source code. | | `find_references` | Find all usages of a symbol. Supports multiple symbols at once. `limit` caps results. | | `get_file_symbols` | List all symbols in a file. Supports multiple files (batch). `include_source=true` to include each symbol's source code. | | `index_directory` | Index a directory (or multiple). Defaults to `workspace_path` if not specified. | | `analyze_code` | Code metrics analysis: file sizes, function counts, long methods, average/max method length. `language` filter. `threshold` sets method length limit (default 50 lines). | | `get_stats` | Index statistics: document count, size in bytes. | | `list_directory` | List directory contents (dirs first, sorted by name). `include_hidden` to show dotfiles. `max_depth` sets recursion depth. | ## Quick Start ```bash # Start OpenSearch docker-compose up -d # Build tools go build -o indexer ./cmd/indexer go build -o mcp-server ./cmd/mcp-server # Index a project via CLI ./indexer index /path/to/project --settings settings.json # Or run the MCP server (auto-indexes workspace on startup) ./mcp-server --settings settings.json --transport stdio ``` ## Indexing a Specific Directory ### 1. Via settings.json Set `workspace_path` to your target project: ```json { "workspace_path": "/home/user/my-project", "indexed_extensions": [".go", ".ts", ".py"], "auto_index_on_start": true } ``` Then start the server — it will index the directory automatically: ```bash ./mcp-server --settings settings.json --transport stdio ``` ### 2. Via CLI indexer (any directory, no config needed) ```bash # Index a single directory ./indexer index /home/user/my-project --settings settings.json # Index multiple directories ./indexer index /home/user/project-a --settings settings.json ./indexer index /home/user/project-b --settings settings.json ``` ### 3. Via Docker Compose (mount your project) Edit `docker-compose.yml` to mount your target project: ```yaml services: mcp-server: volumes: - /home/user/my-project:/workspace # mount your project - ./settings.docker.json:/app/settings.json ``` ```bash docker-compose up -d ``` The server will auto-index `/workspace` on startup. ### 4. Via MCP tool `index_directory` at runtime After the server is running, call the `index_directory` tool with any path: ``` index_directory(path: "/home/user/my-project") ``` This works over both stdio and HTTP transport and does **not** require a server restart. ## Architecture - **Indexer** (`cmd/indexer`) — scans files, parses with tree-sitter, stores in OpenSearch - **File Tracker** (`internal/filetracker`) — SQLite for incremental indexing (hash + size tracking) - **MCP Server** (`cmd/mcp-server`) — MCP server exposing search & index tools ## Project Structure ``` cmd/ indexer/ # CLI indexer mcp-server/ # MCP server for AI agents internal/ config/ # settings.json loader filetracker/ # SQLite file tracking scanner/ # filesystem scanner parser/ # tree-sitter Go parser opensearch/ # OpenSearch client mcp/ # MCP server implementation ``` ## Configuration See `settings.json`: - `indexed_extensions` — file types to index (`.go`) - `exclude_patterns` — directories/files to skip - `workspace_path` — project root to auto-index (`.` resolves to settings.json dir) - `auto_index_on_start` — whether to index workspace on MCP server startup - `opensearch` — connection settings ## MCP Server ### Running ```bash # stdio (for Claude Desktop, etc.) ./mcp-server --transport stdio --settings settings.json # HTTP (with health endpoint) ./mcp-server --transport http --port 3000 --settings settings.json ``` ### Docker Compose ```bash docker-compose up -d ``` - OpenSearch: http://localhost:20900 - MCP HTTP: http://localhost:3000 - Health: http://localhost:3000/health The current directory is mounted as `/workspace` inside the container. ## TODO - [ ] Test indexing on non-`.go` extensions (`.ts`, `.py`, `.js`, `.rs`) - [ ] Drop-in file watcher — `watch --dirs ./project-a,./project-b` — no Docker needed, just point at folders and go - [ ] Multi-project support — watch several repos simultaneously, keep all indices in one OpenSearch instance