Obris logo

Local MCP Server

Run the Obris MCP server on your machine using an API key. This connects your AI tools to your Obris topics over stdio without going through the hosted server.

This is an alternative to the remote MCP connectors which use OAuth and connect directly to mcp.obris.ai. The local server is useful when you want to run the server yourself or when your AI tool doesn't support remote MCP connections.

Before you start

To set up the local Obris MCP server, you'll need:

  • An Obris account
  • An API key
  • uv installed
  • A supported AI platform: Claude Desktop, Claude Code, or Gemini CLI

Set your API key

Add your Obris API key to your shell profile so it's available in every terminal session. Add the following line to your ~/.zshrc, ~/.zprofile, or ~/.bash_profile:

export OBRIS_API_KEY=your_api_key_here

Replace your_api_key_here with your Obris API key. Then reload your shell:

source ~/.zshrc

ChatGPT does not support local stdio MCP servers. Use the remote MCP connector or a Custom GPT instead.

Connect your AI platform

Select your AI platform below to see the setup instructions.

Connect via the Claude Desktop app config.

  1. In Claude Desktop, go to Settings > Developer and click Edit Config to open claude_desktop_config.json
  2. Run which uvx in your terminal to get the full path to uvx
  3. Add the Obris server to your config:
    {
  "mcpServers": {
    "obris": {
      "command": "/full/path/to/uvx",
      "args": ["obris-mcp"],
      "env": {
        "OBRIS_API_KEY": "$OBRIS_API_KEY"
      }
    }
  }
}
  4. The config above reads OBRIS_API_KEY from your shell environment (see prerequisites)
  5. Restart Claude Desktop

Check the connection

Once connected, verify that Obris is working by asking your AI assistant a question about your topics. Try a prompt like:

What topics do I have in Obris?

Your assistant should be able to list your Obris topics and the knowledge saved in them.

AI tools decide when to invoke MCP tools. If your assistant doesn't use your Obris data, try mentioning Obris explicitly, for example "Using Obris, find my knowledge about..."