Do I need to run a separate process for either server?

For Milvus, yes — its server is a Python process you run and point at a Milvus URI. Weaviate's MCP server is built into Weaviate itself and reached over a remote endpoint with a bearer token, so there is no separate server process to manage.

Which offers more search variations?

Milvus's server exposes more distinct search tools — vector, full-text, hybrid, and text-similarity search — plus broad collection management. Weaviate's built-in server centers on hybrid vector-and-keyword search with schema and tenant inspection and object upserts, a more compact surface.

Weaviate vs Milvus

Weaviate MCP and Milvus MCP are both official servers for popular open-source vector databases, but they differ in how they connect and how rich their toolset is. Weaviate's is a built-in MCP server: it is exposed by Weaviate itself over a remote endpoint with a bearer token, and gives an agent hybrid vector-and-keyword search, schema (collection config) inspection, tenant listing, and object upserts — a compact set of core operations served directly by the database. Milvus's server, from Zilliz, runs as a Python process pointed at a Milvus URI and offers a broader toolset: vector search, full-text search, hybrid search, text-similarity search, arbitrary queries, and full collection management (list, create, get info, load, and more). So Weaviate's server is a lean, database-native surface reached remotely, while Milvus's is a fuller, locally-run toolkit with more search variations and collection control. Here is a balanced look at how they differ.

How they compare

Dimension	Weaviate	Milvus
How it's served	Built into Weaviate itself, reached over a remote endpoint with a bearer token — no separate process to run.	A separate Python server (from Zilliz) you run and point at a Milvus URI.
Search modes	Hybrid vector-and-keyword search as the core query path, plus schema and tenant inspection.	Vector, full-text, hybrid, and text-similarity search — several distinct query tools.
Write and management	Object upserts plus collection config and tenant listing — a focused management surface.	Broad collection management: list, create, get info, load, and arbitrary queries alongside writes.
Tool breadth	Compact, database-native set of tools tuned to Weaviate's hybrid-search model.	Wider toolkit with more search variations and explicit collection lifecycle operations.
Best-fit task	Hybrid search and upserts against a Weaviate database using its own built-in, remotely-reachable server.	Working across many Milvus collections with multiple search modes and full collection management from a local server.

Verdict

Both are official servers for capable open-source vector databases, so choose by deployment style and how much breadth you need. Pick Weaviate's built-in server when you want a lean, database-native surface reached remotely with a bearer token — hybrid vector-and-keyword search, schema and tenant inspection, and object upserts — with no separate process to manage. Pick the Milvus server when you want a broader toolkit run locally against a Milvus URI: vector, full-text, hybrid, and text-similarity search plus full collection management. In short: Weaviate for a compact, built-in hybrid-search surface; Milvus for a wider, locally-run set of search modes and collection controls.

FAQ

Do I need to run a separate process for either server?: For Milvus, yes — its server is a Python process you run and point at a Milvus URI. Weaviate's MCP server is built into Weaviate itself and reached over a remote endpoint with a bearer token, so there is no separate server process to manage.
Which offers more search variations?: Milvus's server exposes more distinct search tools — vector, full-text, hybrid, and text-similarity search — plus broad collection management. Weaviate's built-in server centers on hybrid vector-and-keyword search with schema and tenant inspection and object upserts, a more compact surface.