# Let an agent query an index through MCP without writing code

> Let an agent query an index through MCP without writing code: 1 companies fully do it, 10 with a limit, 18 carry the tag. Every listing in Retrieval & vector search with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/vector-search.agent-query-tool/
- Markdown: https://www.anchorterminal.com/niches/vector-search.agent-query-tool/index.md (~1,300 tokens)
- Slim: https://www.anchorterminal.com/niches/vector-search.agent-query-tool/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/vector-search.agent-query-tool/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

One company's docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 1 | 10 | 18 | 6 | 5 |

## What counts as done

The agent lists collections and their fields, sends a text query with a filter on a real field, and gets results, without building a vector or guessing a field name, using read-only credentials.

Measured, for our own tests: the first useful result arrives within three tool calls.

How agents ask for it: search the vector store for X; what collections are there; find records where status is open about Y

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **Weaviate API + MCP (Weaviate B.V.)**: weaviate-collections-get-config shows property names, weaviate-tenants-list lists tenants, and weaviate-query-hybrid takes text and filters; a viewer-role key or the read-only switch keeps it read-only.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Chroma API + MCP (Chroma Inc.)**: chroma-mcp lists collections with paging, returns collection info and runs text queries with filters, but it has no read-only mode, so read-only credentials depend on the client used.
- **Cloudflare AI Search (Cloudflare, Inc.)**: The per-instance MCP endpoint exposes one search tool, so the agent cannot list instances or fields, and the endpoint is public once enabled unless something like Cloudflare Access sits in front.
- **Elasticsearch (Elasticsearch B.V.)**: The Agent Builder MCP server exposes built-in and custom tools, limited by the authorising user's permissions, but it needs Kibana Agent Builder (Elastic Cloud or 9.2 and later) and some built-in tools are unavailable over MCP.
- **Meilisearch (Meili SAS)**: The built-in MCP route has four read-only tools, including searchInIndexes with filters and hybrid, but it is experimental and off by default, and describeIndex returns five sample documents rather than a field list.
- **Milvus and Zilliz Cloud API + MCP (Zilliz Inc.)**: The MCP server lists collections, returns schema with milvus_get_collection_info and runs text search with filter_expr, but it has no read-only mode and also has create, load, release and delete tools.
- **OpenSearch (OpenSearch Software Foundation, a project of The Linux Foundation)**: Core MCP tools list indices, return mappings and run DSL searches, but the GenericOpenSearchApiTool can call any endpoint, #43 shows a wrong field returning an empty result the model trusted, and #203 filters ignored in multi mode.
- **Pinecone API + MCP (Pinecone Systems Inc.)**: MCP tools list and describe indexes and search with text and metadata filters, but only on legacy integrated-embedding indexes, and no read-only mode was found.
- **Qdrant API + MCP (Qdrant Solutions GmbH)**: The MCP server takes text queries and has QDRANT_READ_ONLY, but its two tools cannot list collections or fields, and filters need extra configuration.
- **Typesense API + MCP (Typesense, Inc.)**: The Typesense Cloud MCP server runs keyword and hybrid searches with permissions chosen at connect, but first connect is a browser consent page, so a headless agent cannot start alone.
- **Upstash Vector API + MCP (Upstash, Inc.)**: The hosted MCP server lists indexes and namespaces, runs vector_query with server-side embedding, and a read-only OAuth grant or key removes write tools, but no tool lists metadata field names.

- not this job: Epsilla Vector Database (Epsilla, Inc.): No MCP server found in the docs index.
- not this job: LanceDB (LanceDB Systems, Inc.): No official MCP server or hosted endpoint for an agent to call without an Enterprise deployment.
- not this job: Vespa (Vespa.ai AS): No official public MCP server; an MCP endpoint must be built into the user's own application.

## What stops agents finishing

- data tools need a browser sign-in so a headless agent gets cluster tools only
- MCP server lags the engine (no hybrid, no filters without extra flags)

## Around it

After: vector-search.rag-over-docs.

## Where this comes from

Weight 6 of 100 in Retrieval & vector search: 5 of 37 accounts on MCP servers (Milvus, OpenSearch, Meilisearch, Chroma, Pinecone)

As data: https://www.anchorterminal.com/api/v1/use-cases/vector-search.agent-query-tool.json, or the MCP tool get_use_case with id vector-search.agent-query-tool. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
