The zilliztech/mcp-server-milvus project (documented at milvus.io/docs/milvus_and_mcp.md) exposes Milvus vector-database operations to MCP clients such as Claude Desktop and Cursor. The recommended launch path is `uv run src/mcp_server_milvus/server.py --milvus-uri http://localhost:19530` without a separate install step, with optional `MILVUS_URI`, `MILVUS_TOKEN`, and `MILVUS_DB` environment variables. Tools listed in Milvus docs include `milvus-text-search`, `milvus-hybrid-search`, `milvus-multi-vector-search`, `milvus-query`, and `milvus-count` for collection management, semantic retrieval, filtered hybrid search, and entity counts.
Use cases
- Let coding agents run hybrid vector+metadata searches against Milvus collections
- Debug RAG pipelines by querying and counting entities without hand-written SDK scripts
- Prototype multi-vector retrieval flows from Cursor using documented MCP tools
- Connect local Milvus standalone or Zilliz Cloud URIs via MILVUS_URI configuration
- Manage collections through MCP instead of ad-hoc Python notebooks
Key features
- Claude Desktop
- Cursor
Frequently Asked Questions
- Do I need a running Milvus instance?
- Yes. Docs configure `--milvus-uri` or `MILVUS_URI` against local standalone, Docker, or Zilliz Cloud endpoints.
- Is uv required?
- The README recommends running directly with uv—the same pattern shown for Claude Desktop and Cursor samples.
- How does this differ from Pinecone MCP?
- This server targets Milvus/Zilliz stacks and documents Milvus-specific hybrid and multi-vector tools rather than Pinecone indexes.
Related
Related
3 Indexed items
ClickHouse MCP Server
The open-source ClickHouse MCP server (PyPI package `mcp-clickhouse`, repository ClickHouse/mcp-clickhouse) exposes MCP tools such as `run_query`, `list_databases`, and paginated `list_tables` against ClickHouse clusters, defaulting to read-only SQL unless `CLICKHOUSE_ALLOW_WRITE_ACCESS` is enabled. Optional chDB extras add `run_chdb_select_query` for embedded queries over files and URLs. HTTP/SSE transports require authentication via `CLICKHOUSE_MCP_AUTH_TOKEN`, FastMCP OAuth/OIDC providers, or explicit `CLICKHOUSE_MCP_AUTH_DISABLED=true` for local dev; a `/health` endpoint supports orchestrator probes without credentials per README guidance.
Weaviate MCP Server
Weaviate documents a built-in Model Context Protocol server in the main `weaviate/weaviate` binary from v1.37.1 onward at docs.weaviate.io/weaviate/mcp/mcp-server, exposed as a Streamable HTTP endpoint at `/v1/mcp` on the same port as the REST API (default 8080). Enable with `MCP_SERVER_ENABLED=true`; optional `MCP_SERVER_WRITE_ACCESS_ENABLED=true` registers `weaviate-objects-upsert`. Tools include `weaviate-collections-get-config`, `weaviate-tenants-list`, `weaviate-query-hybrid`, and `weaviate-objects-upsert` (write-gated). Authentication uses existing API keys/Bearer tokens with RBAC permissions `read_mcp`, `create_mcp`, and `update_mcp` per Weaviate 1.37 release notes. The standalone weaviate/mcp-server-weaviate repository is deprecated in favor of this built-in server.
Chroma MCP Server
Chroma documents an official Model Context Protocol server in the chroma-core/chroma-mcp repository and docs.trychroma.com/integrations/frameworks/anthropic-mcp, started with `uvx chroma-mcp` over stdio. Tools include `chroma_list_collections`, `chroma_create_collection`, `chroma_peek_collection`, `chroma_modify_collection`, `chroma_delete_collection`, `chroma_add_documents`, `chroma_query_documents`, `chroma_get_documents`, `chroma_update_documents`, and `chroma_delete_documents` per the README. Client types documented: ephemeral (default), persistent (`--client-type persistent --data-dir`), HTTP self-hosted, and Chroma Cloud (`--client-type cloud` with API keys). Embedding function options include default, Cohere, OpenAI, Jina, VoyageAI, and Roboflow per Chroma MCP docs.