MCP server · auth · triggers

Give your agent GitHub, Slack, Stripe and 60 more apps through one MCP URL

Your users connect each app on a page we host. We keep their tokens encrypted, refresh them and make the calls. The agent sees tools, never a credential.

.mcp.json
{
  "mcpServers": {
    "prompt-tm": {
      "type": "http",
      "url": "https://www.prompt-tm.com/mcp/mcp_k3x9…?user_id=alice",
      "headers": { "x-api-key": "ptm_…" }
    }
  }
}
agent ↔ prompt-tm
> tools/call GITHUB_CREATE_ISSUE
  { "owner": "acme", "repo": "web", "title": "Checkout 500 on Safari" }

< { "successful": true,
    "data": { "number": 128, "html_url": "https://github.com/acme/web/issues/128" },
    "log_id": "log_7f2…" }
Works withClaude CodeClaude Desktopclaude.aiCursorVS CodeWindsurfOpenAI Agents SDKAny MCP client
toolkits
64
tools
935
triggers
13
MCP URL
1

How it works

How an agent gets its first tool

  1. 01

    Pick the apps

    Add an auth config for each app. Choose your own OAuth client, Prompt TM's app where we offer one, or a key form, and limit the scopes and tools if you like.

  2. 02

    Send your user a connect link

    The user signs in to the app, or pastes a key, on prompt-tm.com. Your app gets the connected account id back.

  3. 03

    Point the agent at the MCP URL

    Paste one URL into Claude, Cursor or your own agent. Every call it makes runs with that user's account.

Platform

What you would otherwise build for every app

Auth management on our side

OAuth 2.0 with PKCE, API keys, access tokens and basic auth. We encrypt every token with AES-256-GCM. A call that finds its token about to expire refreshes it first. If a provider refuses, the account turns EXPIRED and your webhook hears about it.

A router for open-ended agents

The router lists six tools instead of hundreds. Your agent searches the catalogue, asks the user to connect a missing app, runs up to 20 tools at once, and works through large results in a code sandbox.

Triggers that register themselves

Switch on a GitHub or Stripe trigger and we create the webhook in your user's account. We check each delivery's signature and forward the event to your URL, signed with your secret.

A log line for every call

Each call records the tool, user, account and duration, plus a trimmed copy of the request and response with secrets masked. Filter by app, tool, user or status.

REST when you don't want MCP

Run any tool with one POST, or fetch the tool list already shaped for the OpenAI or Anthropic SDK. For anything a tool doesn't cover, the proxy endpoint calls the app's API with the user's auth.

Many users, many accounts

Each user is an id you choose. One user can connect several accounts of the same app, name them, and you pick which one a call uses.

REST API

The same tools without MCP

Every tool is one POST, and every answer has the same four fields. Your agent loop treats all apps alike. Ask for the tool list with format=anthropic or format=openai and pass it straight to the SDK.

API reference
terminal
curl -X POST https://www.prompt-tm.com/api/v3/tools/execute/VERCEL_LIST_DEPLOYMENTS \
  -H "x-api-key: $PROMPT_TM_KEY" \
  -H "content-type: application/json" \
  -d '{ "user_id": "alice", "arguments": { "limit": 5, "target": "production" } }'

FAQ

Before you connect anything

Connect GitHub and make your agent's first tool call

Create a project, paste a GitHub token, add the MCP URL to Claude Code.

Create a free project