Overview

The Meilisearch MCP Server implements the Model Context Protocol (MCP) so language models and agent frameworks can manage Meilisearch instances through conversational flows. It exposes Meilisearch functionality as MCP tools over stdio, letting any MCP-compatible client (Claude, OpenAI agents, or other LLM-driven systems) create indices, add documents, run searches, tune settings, and monitor tasks without calling the Meilisearch HTTP API directly.

This server is useful when you want to embed search and retrieval into agent workflows or conversational applications. Instead of embedding the Meilisearch HTTP client inside your agent logic, the agent talks to the MCP server; the server translates agent actions into Meilisearch operations, handles credentials, and returns structured results suitable for downstream reasoning or presentation.

Repository: https://github.com/meilisearch/meilisearch-mcp

Features

Index management: create, update, delete indices and primary keys
Document operations: add, update, and delete documents (batching supported)
Search and multi-index search: natural and filtered searches, sorting, pagination
Settings management: update ranking rules, synonyms, stop words, searchable/sortable attributes
Task and health monitoring: track asynchronous indexing tasks and instance health
API key handling: configure and switch API keys securely at runtime
Dynamic connection switching: connect to multiple Meilisearch instances without restarting
stdio MCP transport: compatible with MCP clients that use stdio (e.g., Claude Desktop); other transports may be added later
Deployable via PyPI package, Docker image, or from-source installation

Installation / Configuration

Prerequisites:

Python ≥ 3.9 (for source / pip installation) or Docker
A running Meilisearch instance
MCP-compatible client (Claude Desktop, OpenAI agents, etc.)

Install from PyPI:

pip install meilisearch-mcp

Run with uvx (recommended for MCP stdio integration):

uvx -n meilisearch-mcp

Run via Docker (recommended in containers and CI):

docker run -it \
  -e MEILI_HTTP_ADDR=http://your-meilisearch:7700 \
  -e MEILI_MASTER_KEY=your-master-key \
  getmeili/meilisearch-mcp:latest

Common environment variables:

Variable	Purpose	Example
MEILI_HTTP_ADDR	Meilisearch HTTP address	http://meilisearch:7700
MEILI_MASTER_KEY	Master API key for Meilisearch	masterKey
MCP_LOG_LEVEL	Logging level for the server	info, debug

Configure Claude Desktop (example snippet):

{
  "mcpServers": {
    "meilisearch": {
      "command": "uvx",
      "args": ["-n", "meilisearch-mcp"]
    }
  }
}

From source (development):

git clone https://github.com/meilisearch/meilisearch-mcp.git
cd meilisearch-mcp
uv venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
uv pip install -e .
uvx -n meilisearch-mcp

Available Tools / Resources

The server exposes a set of MCP tools that agents can call. Typical tool names and capabilities include:

list_indices — List existing indices and basic stats
create_index — Create an index with optional primary key
delete_index — Remove an index
add_documents — Add or replace documents in an index (batch)
get_document / delete_document — Single-document operations
search — Query an index with filters, sort, and pagination
search_multi — Run a search across multiple indices and aggregate results
update_settings — Modify ranking rules, synonyms, stop words, etc.
get_tasks — Inspect indexing tasks and status
health — Check Meilisearch instance health
api_keys — List and manage API keys (if enabled)

Return payloads are JSON-serializable and structured to make it easy for an LLM to reason over results (e.g., top hits, counts, task statuses).

Use Cases

Conversational RAG (Retrieval-Augmented Generation): Connect an assistant to Meilisearch for retrieving relevant documents (e.g., knowledge base articles) in response to user queries, then synthesize answers with contextual citations.
Agent-driven data management: Let an autonomous agent create indices, ingest data, and tune searching behavior as part of a workflow (e.g., automating e-commerce catalog updates).
Multi-index exploration: Ask an assistant to search across several indices (blogs, docs, tutorials) and summarize cross-index findings.
Integrations with automation platforms (n8n, Airflow): Run the MCP server as a container to expose search capabilities to low-code flows and orchestrations.

Example conversation-driven flow:

User: “Create an index called products with id as primary key.”
Agent calls create_index({“name”:“products”,“primaryKey”:“id”})
Server returns success and index metadata.
User: “Add a few products.” Agent calls add_documents with a list of product JSONs.
User: “Search for electronics under $50.” Agent calls search with filter “category = ‘electronics’ AND price <= 50”.

Tips and Best Practices

Use sortable attributes for fields you plan to sort on (e.g., “date”, “price”).
Batch document uploads to reduce task overhead.
Use multi-index searches for cross-collection discovery, then refine with index-specific filters.
Manage API keys and environment variables securely in production deployments.

For full implementation details, supported tool signatures, and examples, see the repository: https://github.com/meilisearch/meilisearch-mcp

Meilisearch MCP Server for Conversational Search