Skip to main content
The Model Context Protocol (MCP) lets AI assistants like Claude interact directly with Meilisearch. Once configured, you can create indexes, add documents, configure settings, and perform searches using natural language prompts.

​
Requirements

To follow this guide, you’ll need:

​
Setting up Claude Desktop with the Meilisearch MCP Server

​
1. Install Claude Desktop

Download and install Claude Desktop.

​
2. Install the Meilisearch MCP Server

You can install the Meilisearch MCP server using uv or pip:

​
3. Configure Claude Desktop

Open Claude Desktop, click on the Claude menu in the top bar, and select “Settings”. In the Settings window, click on “Developer” in the left sidebar, then click “Edit Config”. This will open your claude_desktop_config.json file. Add the Meilisearch MCP server to your configuration:
Save the file and restart Claude.

​
Connecting to Your Meilisearch Instance

Once Claude Desktop is set up with the Meilisearch MCP server, you can connect to your Meilisearch instance by asking Claude to update the connection settings. Open Claude Desktop and start a new conversation. Next, connect to your Meilisearch instance by asking Claude to update the connection settings, replacing MEILISEARCH_URL with your project URL and API_KEY with your project’s API key:
Claude will use the MCP server’s update-connection-settings tool to establish a connection to your Meilisearch instance. Finally, verify the connection by asking:
Claude will use the get-version and health-check tools to verify the connection and provide information about your instance.

​
Create an e-commerce index

Now you have configured the MCP to work with Meilisearch, you can use it to manage your indexes. First, verify what indexes you have in your project:
Next, ask Claude to create an index optimized for e-commerce:
Finally, check the index has been created successfully and is completely empty:

​
Add documents to your new index

Ask Calude to add a couple of test documents to your “products” index:
Since you are only using “products” for testing, you can also ask Claude to automatically populate it with placeholder data:
To verify data insertion worked as expected, retrieve the first few documents in your index:

​
Configure your index

Before performing your first search, set a few index settings to ensure relevant results. Ask Claude to prioritize exact word matches over multiple partial matches:
It’s also a good practice to limit searchable attributes only to highly-relevant fields, and only return attributes you are going to display in your search interface:

​
Perform searches with MCP

Perform your first search with the following prompt:
You can also request your search uses other Meilisearch features such as filters and sorting:

​
Important note about LLM limitation

Large Language Models like Claude tend to say “yes” to most requests, even if they can’t actually perform them.Claude can only perform actions that are exposed through the Meilisearch API and implemented in the MCP server. If you’re unsure whether a particular operation is possible, refer to the Meilisearch documentation and the MCP server README.

​
Troubleshooting

If you encounter issues with the Meilisearch MCP integration, try these steps

​
1. Ask Claude to verify your connection settings

​
2. Ask Claude to check your Meilisearch instance health

​
3. Review Claude’s logs

Open the logs file in your text editor or log viewer:
  • On macOS: ~/Library/Logs/Claude/mcp*.log
  • On Windows: %APPDATA%\Claude\logs\mcp*.log

​
4. Test the MCP server independently

Open your terminal and query the MCP Inspector with npx:

​
Conclusion

The Meilisearch MCP integration with Claude can transform multiple API calls and configuration tasks into conversational requests. This can help you focus more on building your application and less on implementation details. For more information about advanced configurations and capabilities, refer to the Meilisearch documentation and the Meilisearch MCP server repository.