Qdrant

### 1. Basic Configuration This is a minimal, local Qdrant MCP server configuration using an API key from your environment. ```json { "qdrant": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-qdrant" ], "env": { "QDRANT_URL": "http://localhost:6333", "QDRANT_API_KEY": "${QDRANT_API_KEY}" }, "disabled": false } } ``` **Explanation (not part of JSON):** - `"qdrant"`: The MCP server name as it will appear in Claude Code. - `"command": "npx"`: Uses `npx` to run the Qdrant MCP server package without a global install. - `"args"`: - `-y`: Auto‑confirm any prompts. - `@modelcontextprotocol/server-qdrant`: The MCP Qdrant server package. - `"env"`: - `QDRANT_URL`: Points to a local Qdrant instance (default Docker port). - `QDRANT_API_KEY`: Pulled from your shell environment; set it with `export QDRANT_API_KEY=...`. - `"disabled": false`: Ensures the server is active. --- ### 2. Advanced Configuration Adds explicit collection defaults and tuning parameters via environment variables, suitable for a more complete dev setup. ```json { "qdrant": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-qdrant" ], "env": { "QDRANT_URL": "https://qdrant.dev.internal:6333", "QDRANT_API_KEY": "${QDRANT_API_KEY}", "QDRANT_COLLECTION": "claude_context", "QDRANT_VECTOR_SIZE": "1536", "QDRANT_DISTANCE": "Cosine", "QDRANT_TIMEOUT_MS": "10000", "QDRANT_BATCH_SIZE": "128", "QDRANT_TLS_INSECURE": "false" }, "disabled": false } } ``` **Explanation (not part of JSON):** - `"QDRANT_URL"`: Points to a remote/dev Qdrant instance (e.g., in a cluster or remote VM). - `"QDRANT_COLLECTION"`: Default collection name for all operations (created if missing by the server, depending on implementation). - `"QDRANT_VECTOR_SIZE"`: Dimensionality of vectors (e.g., 1536 for many OpenAI-like embeddings). - `"QDRANT_DISTANCE"`: Similarity metric (`Cosine`, `Dot`, or `Euclid` depending on your Qdrant setup). - `"QDRANT_TIMEOUT_MS"`: HTTP timeout for Qdrant requests in milliseconds. - `"QDRANT_BATCH_SIZE"`: How many points to upsert per batch; tune for throughput vs. memory. - `"QDRANT_TLS_INSECURE"`: When `false`, TLS certificates must be valid; set to `true` only for testing self‑signed certs. --- ### 3. Use Case Specific Configuration **Use case: Production semantic search over documentation with separate collections per environment** This configuration assumes: - A managed Qdrant Cloud instance. - Separate collections for `prod` and `staging`. - Stronger timeouts and a read‑only default mode for safety in production. ```json { "qdrant": { "command": "npx", "args": [ "-y", "@modelcontextprotocol/server-qdrant" ], "env": { "QDRANT_URL": "https://your-cluster-1234.us-east-1-0.aws.cloud.qdrant.io", "QDRANT_API_KEY": "${QDRANT_API_KEY}", "QDRANT_COLLECTION": "docs_prod", "QDRANT_VECTOR_SIZE": "3072", "QDRANT_DISTANCE": "Cosine", "QDRANT_TIMEOUT_MS": "20000", "QDRANT_BATCH_SIZE": "256", "QDRANT_TLS_INSECURE": "false", "QDRANT_ENV": "production", "QDRANT_READ_ONLY": "true", "QDRANT_ALT_COLLECTION_STAGING": "docs_staging" }, "disabled": false } } ``` **Explanation (not part of JSON):** - `"QDRANT_URL"`: Managed Qdrant Cloud endpoint for production. - `"QDRANT_COLLECTION": "docs_prod"`: Main collection for production docs semantic search. - `"QDRANT_ALT_COLLECTION_STAGING"`: Optional second collection name for staging docs; the MCP server can expose tools to switch or query both. - `"QDRANT_VECTOR_SIZE": "3072"`: Larger vectors (e.g., from higher‑capacity embedding models). - `"QDRANT_TIMEOUT_MS": "20000"` and `"QDRANT_BATCH_SIZE": "256"`: Tuned for higher latency networks and bulk indexing. - `"QDRANT_ENV": "production"`: Lets the MCP server adjust behavior (logging, safety checks, etc.). - `"QDRANT_READ_ONLY": "true"`: Encourages the server to expose only read/query operations to Claude by default, reducing the risk of accidental destructive changes in production.