Skip to content

Connect Claude / Cursor

  • Floci Studio stack running (docker compose up or native)
  • uv installed (curl -LsSf https://astral.sh/uv/install.sh | sh)

Open your Claude Desktop config file:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the floci entry under mcpServers:

{
"mcpServers": {
"floci": {
"command": "uv",
"args": [
"run",
"--project",
"/path/to/floci-studio/mcp",
"python",
"/path/to/floci-studio/mcp/floci_mcp.py"
],
"env": {
"SIDECAR_TOKEN": "open",
"AWS_ENDPOINT_URL": "http://127.0.0.1:4566"
}
}
}
}

Replace /path/to/floci-studio with the absolute path to your clone.

Restart Claude Desktop. The floci server appears in the MCP panel (hammer icon).

Open Cursor Settings → MCP → Add Server:

{
"floci": {
"command": "uv",
"args": [
"run", "--project", "/path/to/floci-studio/mcp",
"python", "/path/to/floci-studio/mcp/floci_mcp.py"
],
"env": {
"SIDECAR_TOKEN": "open",
"AWS_ENDPOINT_URL": "http://127.0.0.1:4566"
}
}
}

If you prefer not to install Python locally, run the MCP server as a Docker container:

Terminal window
docker build -t floci-mcp /path/to/floci-studio/mcp/
# Claude Desktop config
{
"mcpServers": {
"floci": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "AWS_ENDPOINT_URL=http://host.docker.internal:4566",
"-e", "SIDECAR_TOKEN=open",
"floci-mcp"
]
}
}
}

In Claude Desktop, open a new conversation and type:

“List all SQS queues in floci”

You should see the agent call list_sqs_queues and return the queue list (or an empty array if none exist yet).

“Server disconnected” or no tools appear

  • Make sure uv is on your PATH. Run uv --version in a terminal.
  • Check the absolute path in args is correct.
  • Restart Claude Desktop fully after editing the config.

“Connection refused” on AWS calls

  • Confirm floci engine is running: curl http://localhost:4566/_localstack/health
  • If using Docker transport, the endpoint must be host.docker.internal:4566, not localhost:4566.

Wrong AWS_ENDPOINT_URL

  • The sidecar and MCP server both use AWS_ENDPOINT_URL. Make sure it points to where the engine is listening.