← Back to Overview

MCP SERVERS

Connecting AI Clients

How to connect any MCP-compatible AI client to API Studio's Management MCP server — Claude Desktop, Cursor, Kiro, VS Code Copilot, Amazon Q, and headless CLI environments.

Overview

Any MCP-compatible AI client can connect to API Studio's MCP servers. The Management MCP server exposes 33 tools for full workspace control — collections, environments, requests, mock servers, and more.

Connection URL

http://127.0.0.1:{port}/mcp — default Management MCP port is 3199

Transport

Supports both HTTP (streamable) and stdio transport. Use HTTP for most clients; stdio for environments that prefer subprocess communication.

MCP Connect Panel

The MCP Connect panel in the UI shows connection instructions for each client — including the URL, config snippets, and the stable stdio script path.

Prerequisites: Enable the Management MCP server in VS Code Settings → Extensions → API Studio → Management MCP Enabled. The server must be running for clients to connect.

Recommendation: Use stdio over HTTP for stability

The HTTP transport (http://127.0.0.1:3199/mcp) can be unstable when:

  • Multiple VS Code windows are open (only the first window owns the port)
  • Another process is already using port 3199
  • VS Code restarts or the extension reloads (brief connection gap)

stdio is more reliable — it launches a dedicated subprocess per AI client, avoids port conflicts, and works even when the extension hasn't started yet. Use the stable script path:

/usr/local/bin/node ~/.openpost/global/mcp-stdio.js

Use the full path to node (e.g., /usr/local/bin/node on macOS, C:\Program Files\nodejs\node.exe on Windows) for consistent behavior across shells and environments. Find yours with which node or where node.

Claude Desktop

Edit your Claude Desktop configuration file to add API Studio as an MCP server.

Config file location:

macOS: ~/.config/claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

Stdio transport (recommended — more stable):

{
  "mcpServers": {
    "api-studio": {
      "command": "/usr/local/bin/node",
      "args": ["/Users/YOUR_USERNAME/.openpost/global/mcp-stdio.js"]
    }
  }
}

Replace /usr/local/bin/node with output of which node. Replace YOUR_USERNAME with your actual home path.

HTTP transport (alternative — may have port conflicts):

{
  "mcpServers": {
    "api-studio": {
      "url": "http://127.0.0.1:3199/mcp"
    }
  }
}
} }

Restart Claude Desktop after editing the config. Tools will appear in the tool picker automatically.

Cursor

Add API Studio as an MCP server in Cursor's settings.

StepAction
1Open Settings → MCP
2Click Add Server
3Name: api-studio
4Type: URL
5URL: http://127.0.0.1:3199/mcp
6Click Save

Tools are available in Cursor's AI chat and agent mode immediately after saving.

Kiro

Add API Studio as an MCP server in Kiro's settings.

StepAction
1Open Settings → MCP
2Click Add Server
3URL: http://127.0.0.1:3199/mcp

Tools are available for Kiro's AI agent once the server is added.

VS Code Copilot

VS Code Copilot can connect to MCP servers automatically when enabled.

Connects automatically via MCP when enabled in Copilot Chat settings
May require the MCP extension or Copilot Chat MCP support to be enabled
URL: http://127.0.0.1:3199/mcp
Note: Copilot's MCP support is evolving. Check the latest VS Code Copilot documentation for the most up-to-date setup instructions.

Amazon Q

Add API Studio to Amazon Q's MCP server configuration.

StepAction
1Open MCP settings
2Click Add server → select URL
3Enter: http://127.0.0.1:3199/mcp

Tools are discoverable via Amazon Q's tool calling interface.

CLI Serve (Headless)

For CI/CD or headless environments, start the Management MCP server without VS Code using the CLI.

openpost serve
Starts the Management MCP server without VS Code
Listens on configured port (default 3199)
All 33 tools available
Reads .openpost/ from current directory

Useful for:

AI agents in CI pipelines Remote MCP access Automated testing

Testing the Connection

Verify your MCP connection is working with a quick curl test:

curl -X POST http://127.0.0.1:3199/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

Should return JSON with all 33 tools listed in the result.tools array.

Connection refused? Ensure the Management MCP server is enabled and running. Check the Mock + MCP panel in API Studio — the Management MCP entry should show a green "Running" status.
MCP connection panel showing stdio configuration

MCP connection panel showing stdio configuration

Next

Management MCP → CLI Serve →
Ko-fi