Collection MCP

Overview

Collection MCP servers expose your saved API requests as tools for AI assistants. Any MCP-compatible client can discover and call your endpoints without manual configuration.

AI client compatibility — Claude Desktop, Cursor, Copilot, Amazon Q, and any MCP-compatible assistant

Read-only — AI clients can invoke requests but cannot modify your collections

One server per collection — each collection can have its own MCP server instance

JSON-RPC 2.0 — served via POST /mcp following the MCP specification

How it works: Start an MCP server for a collection → AI client connects to the endpoint → discovers available tools (your requests) → invokes them with parameters → receives the API response.

Starting a Server

Launch a collection MCP server from the Collection Editor's MCP tab (requires Advanced Mode).

1

Open a collection → click the MCP sub-tab in the Collection Editor

2

Click Start MCP Server — a port is auto-assigned (or configure a specific port)

3

Server starts listening — the URL is displayed: http://127.0.0.1:{port}/mcp

4

Point your AI client to this URL to connect

Option	Description
Port	Auto-assigned by default, or set a specific port number
Auto-start	Server starts automatically when VS Code launches
Persistence	Configs saved in `.openpost/mcp-servers.json`

Tool Generation

Each request in the collection automatically becomes an MCP tool that AI clients can discover and invoke.

Tool name — derived from the request name (slugified, e.g. "Get Users" → get_users)

Tool description — from the mcpDescription field (editable per request)

Full pipeline execution — when called, the request runs through auth, scripts, interpolation, and HTTP send

Response returned — the API response is sent back to the AI client

Note: Requests without an explicit mcpDescription are still exposed as tools — the request name is used as the description.

Input Schema

Define a JSON Schema for each tool's input parameters so AI clients know what arguments the tool accepts.

Auto-generate

Click Generate Schema to automatically create a JSON Schema from the request's params and body fields.

Custom schema

Edit the JSON Schema directly for full control over parameter names, types, and descriptions.

When an AI client calls a tool with parameters, the values are injected into the request — mapped to body fields, query params, or headers as appropriate.

Example mcpInputSchema:

{
  "type": "object",
  "properties": {
    "userId": {
      "type": "string",
      "description": "The user ID to fetch"
    },
    "includeProfile": {
      "type": "boolean",
      "description": "Whether to include profile details"
    }
  },
  "required": ["userId"]
}

Call Logging

Every MCP interaction is logged for debugging and monitoring AI client behavior.

Log Field	Description
Timestamp	When the call was received
Method	`tools/list` or `tools/call`
Tool Name	Which tool was invoked
Status	Success or error

Viewable in the MCP Server detail panel (resizable call history pane)

Helps debug AI client interactions and verify tool invocations

Max log entries configurable per server (prevents unbounded growth)

Configuration

MCP server configs are persisted in .openpost/mcp-servers.json with the following fields:

Field	Type	Description
id	string	Unique identifier
name	string	Display name for the server
collectionId	string	Collection this server exposes
scope	string	Local or global scope
port	number	Port number for the server
autoStart	boolean	Start on VS Code launch
swaggerEnabled	boolean	Serve OpenAPI spec at `/_openapi.json`
maxLogEntries	number	Maximum call log entries to keep

Swagger: When swaggerEnabled is true, the server also serves an auto-generated OpenAPI 3.0 spec at /_openapi.json alongside the MCP endpoint — useful for clients that prefer REST discovery.