Meilisearch MCP Server

The Meilisearch MCP server connects AI assistants to a running Meilisearch instance. Anyone using Meilisearch for search in an online shop, documentation site, or application can use it to create indexes, add documents, test search queries, adjust relevancy settings, and check the health of the system in conversation, without reaching for curl commands or the dashboard. According to the README the server uses the stdio transport, is written in Python, and builds on the official MCP Python SDK. It is licensed under MIT and is kept in the GitHub repository of the meilisearch organization; the repository is not archived and was last updated in September 2026.

What is the Meilisearch MCP server?

Meilisearch is an open-source search engine known for fast, typo-tolerant search. The MCP server is a small intermediary: it runs locally on your machine, talks to your instance over the Meilisearch HTTP API, and exposes its functions to the assistant as tools. The README lists 22 tools in total. They cover the connection, indexes, documents, search, settings, API keys, tasks, and system monitoring. For every tool that returns data, the server also returns structured output alongside the text, according to the README, which makes it easier for clients to process results.

What can the assistant do with it?

Three areas matter most in daily work. First, data upkeep: with create-index, add-documents, and get-documents you can build and inspect test indexes, for example to try out a product catalog. Second, search itself: the search tool queries one or several indexes with filters and sorting, so you can settle ranking questions directly in the chat. Third, settings: get-settings and update-settings let you view and adjust ranking rules and facets. On top of that come get-index-metrics, get-stats, get-version, health-check, and get-health-status for a quick overview, plus get-tasks and cancel-tasks to watch or stop background tasks of the instance.

Special features: dynamic connection and API keys

One peculiarity is switching the connection during a conversation. With get-connection-settings and update-connection-settings the assistant can change the address and the key and so move between a local development instance and a test instance. API keys can also be managed with get-keys, create-key, and delete-key. The README itself points out that adding hosts and keys in chat is primarily meant for development, does not follow best practices for MCP security, and should not be used in production without additional safeguards.

Setup

You need Python 3.10 or newer, a running Meilisearch instance, and an MCP-capable client. The README recommends starting with uvx -n meilisearch-mcp; alternatives are pip install meilisearch-mcp and a Docker image (getmeili/meilisearch-mcp). The target address and the key are set through the environment variables MEILI_HTTP_ADDR (default http://localhost:7700) and MEILI_MASTER_KEY, where the key is optional. In Claude Desktop you register the server in the configuration file; in Claude Code a single command is enough.

Security and limits

The server can not only read but also write: it creates indexes, deletes them together with all documents (delete-index), and creates or removes keys. That makes it powerful and calls for care. For experiments use a key with narrowly scoped permissions instead of the master key, and never point the assistant at a production instance whose data you cannot replace. Documents and search results the assistant retrieves also enter the language model's context and thereby reach the model provider.

Costs

The MCP server itself is free and open source. According to the vendor Meilisearch is also open source under the MIT license and can be self-hosted without a license fee; you then take care of updates, backups, and scaling yourself. If you prefer a hosted service, Meilisearch Cloud offers paid plans with usage-based or resource-based billing and a free trial period, plus an enterprise tier on request. Current terms are on the official pricing page.

Who is the Meilisearch MCP server for?

It is worthwhile for developers who already use or are evaluating Meilisearch: for quickly building test indexes, trying out ranking settings, and answering questions such as "Why does this product not show up at the top?". Teams that let a coding agent build search features into an application also benefit from access to a real development instance, because the agent can check results immediately. Anyone not using Meilisearch gains nothing from it, and according to the README the server is not meant for unattended production use.

Python 3.10 or newer (or Docker), a running Meilisearch instance, and an MCP-capable client; an API key is optional but needed for protected instances.

The Meilisearch MCP server runs locally and controls a Meilisearch instance through its HTTP API. This guide walks through the setup.

Step 1: Prerequisites

You need Python 3.10 or newer (for uvx), a running Meilisearch instance, and an MCP-capable client. According to the README you can start a local instance, for example, with Docker using the getmeili/meilisearch image on port 7700.

Scope the key narrowly

Create a dedicated API key with the necessary permissions for the assistant instead of using the master key. The server can delete indexes and manage keys, so the key limits the possible damage.

Step 2: Register the server in Claude Code

Run claude mcp add meilisearch -e MEILI_HTTP_ADDR=http://localhost:7700 -e MEILI_MASTER_KEY=YOUR_KEY -- uvx -n meilisearch-mcp. Without a key, leave out the second variable. Afterwards check with /mcp that the server is connected.

claude mcp add meilisearch -e MEILI_HTTP_ADDR=http://localhost:7700 -e MEILI_MASTER_KEY=YOUR_KEY -- uvx -n meilisearch-mcp

Step 3: Register the server in Claude Desktop

In claude_desktop_config.json add an entry meilisearch under mcpServers with the command uvx and the arguments -n and meilisearch-mcp. If needed add an env block with the two variables and restart the client.

Alternative: Docker

Instead of uvx you can use the getmeili/meilisearch-mcp image and pass MEILI_HTTP_ADDR and MEILI_MASTER_KEY as environment variables. If Meilisearch runs on the host, the address must be reachable from inside the container.

Step 4: First requests

Ask the assistant to query the version and health of the instance, then to create a test index and add a few documents. Next run search queries with filters and check through the task list whether indexing has finished.

Not for production

The README warns that switching hosts and keys in chat is meant for development. Do not connect the assistant to a production instance whose data you cannot replace.

Through the optional MEILI_MASTER_KEY (or a narrowly scoped API key) of the Meilisearch instance; the server itself has no sign-in of its own.

Read and write access to the connected instance within the scope of the key used: indexes, documents, settings, tasks, and API keys can be created, changed, and deleted.

The server runs locally and talks only to the configured Meilisearch instance. Retrieved documents and search results enter the AI client context and therefore reach the model provider.

Accidentally deleting indexes together with their documents, using the master key instead of a narrowly scoped one, switching the connection in chat to a production instance, and the assistant creating or removing API keys.