OPENAPI MCP

Turn an API into tools an agent can use.

Approve the operations, test their behavior, and hand your agent a small, typed MCP surface. This demo uses public example data.

GET onlyApproved operationsFixed HTTPS originBounded responses

01 / DISCOVER

Approved tools

3 tools

Selected from trusted OpenAPI files. Write operations stay outside the MCP server.

TRUSTED SOURCESJSONPlaceholder posts · GitHub public issues

02 / VERIFY

Run a tool

Live request

RESPONSEReady

Select a tool and run a request to see the response.

03 / DELEGATE

Ask the cloud agent

Checking cloud

Describe the result you need. An OpenAI model on Agent37 selects one approved tool; the bridge validates it and runs the public API request.

AGENT TRACEReady

The plan and verified API response appear here.

04 / GENERATE

Bring your own API

Local file only

Choose an OpenAPI JSON file, review its GET operations, and download a configuration for the MCP server. The file stays in your browser.

Then set BRIDGE_CONFIG_FILE to the downloaded file and run npm start.

05 / CONNECT

A small surface for a capable agent.

The approved operations are discoverable through the Model Context Protocol. The local test suite verifies tool discovery, argument rejection, and bounded upstream responses.

View source and setup