OATTY

Guide

Give your agent access to an API

Connect your existing agent to Oatty through MCP. Import an OpenAPI document, find one operation, and run it. You can then find and run the same command in Oatty's terminal interface.

This guide uses a read operation. Have the API's OpenAPI document or URL, its base URL, and any required credentials ready. Use an API and credentials you are allowed to access.

  1. Install and start Oatty
  2. Connect your MCP client
  3. Import the API
  4. Run one call
  5. Run it yourself

Install and start Oatty

npm install -g oatty
oatty --help
oatty

In the TUI, open the MCP HTTP Server view and start the server. Read its Active endpoint. The default local endpoint in the current documentation is http://localhost:62889/mcp; use the address shown by your running instance if it differs.

For server controls and connection troubleshooting, see the MCP HTTP server guide.

Connect your MCP client

Add the active endpoint as an HTTP MCP server in your client's settings, then reconnect the client. For a client that accepts a server URL, the local setting looks like this:

http://localhost:62889/mcp

Client configuration formats differ. Verify the client can list Oatty tools before continuing. The TUI shows the server status and connected client count.

For authenticated APIs, the inspected source supports request headers backed by raw values, environment variables, files, or secrets. Configure the required header before the first call, and inspect it through the masked-header view. Native OAuth token refresh has not been verified.

Import the API

Give the agent the path or URL of your OpenAPI document. Ask it to call catalog_validate_openapi, then catalog_preview_import. Review the validation result, generated command preview, and chosen base URL. Once those match your API, have it call catalog_import_openapi with a clear catalog title.

Use Oatty's MCP tools to validate and preview this OpenAPI document: <path-or-URL>.
Show me the generated read operation and base URL. Then import it as <catalog-title>.

If the API does not publish an OpenAPI document, follow author a schema from API documentation.

Run one API call

Ask the agent to find a specific read operation with search_commands, inspect its exact inputs with get_command, and execute it with run_safe_command. Supply any required inputs. Check the returned status and response body against the API's expected result.

Find the command that reads <resource> from <catalog-title>.
Inspect its required inputs, run it, and show the response status and body.

Use run_command for a POST, PUT, or PATCH operation and run_destructive_command for DELETE. A tool name does not establish that an operation is harmless; check the API operation before executing it.

Run the same command yourself

Return to Oatty's TUI. Search for the imported operation in the command catalog. Select the command from autocomplete, fill its inputs, and run it. Inspect the response and execution logs. The agent and TUI use the imported catalog, so the command can be examined from either side.

If the call fails, inspect the response and inputs first. For a wrong generated path, method, or connection setting, continue to correct a catalog command.