AgentPlay reads your OpenAPI spec and hosts a live MCP gateway — so Claude, ChatGPT, Gemini, and any MCP-compatible client can call your real API. Zero changes to your existing backend.
Free plan · No credit card required · Live in under 5 minutes
Convert this PDF report to DOCX and email it to the team.
Using convert_file from Acme API…
Using send_email from Acme API…
Done. Converted and sent to the team in 2.1 seconds.
One MCP endpoint works with every major AI client
The problem
Giving them that is harder than it should be.
An MCP server needs auth, transport, tool schema definitions, rate limiting, and hosting. It's a full engineering project before a single user gets value.
Passing API keys through your users' Claude sessions is a security liability. Keys belong in the gateway, not the prompt.
Which tools are agents actually calling? What's failing? Which users are hitting limits? You need answers before something breaks in production.
How it works
Three steps. No infrastructure to manage. No code changes to your existing API.
Paste a URL, upload a file, or use our AI prompt to generate a spec from scratch. Our AI reads your endpoints, writes clean descriptions, and skips destructive operations by default.
You get a unique, authenticated URL. Paste it into Claude, ChatGPT, Gemini, Cursor, or any MCP-compatible client. Tools are discovered instantly — no client code, no SDK, no configuration.
Real-time call logs, per-tool analytics, built-in rate limiting, and AES-256 encrypted credentials. Your upstream API key stays locked in our vault — never visible to end users.
A file-conversion company wants their users to say “convert this PDF to DOCX” to Claude and get a real result back. Normally that's a multi-week engineering project. With AgentPlay, their developer does it between breakfast and lunch.
Try it with your API →Paste OpenAPI spec URL into AgentPlay
AI curates convert_file, list_formats, get_status tools
Add MCP URL as a Claude custom connector
Users call the real API through Claude — done
Features
Security, observability, and reliability — already built. You just connect your spec.
Drop in an OpenAPI URL or paste raw JSON/YAML. Our AI writes tool names, descriptions, and parameter docs — and disables destructive endpoints by default. Don't have a spec? Use our AI prompt to generate one.
Your upstream API key is AES-256-GCM encrypted at rest and decrypted only inside the request handler. It never appears in logs, user sessions, or response payloads.
Rename tools, rewrite descriptions, enable or disable individual endpoints, and edit parameter documentation — all from the dashboard. Edits survive re-imports.
Every upstream URL is validated at config time and again at request time. Private IP ranges, localhost, and link-local addresses are always blocked.
Sliding-window rate limiter per project and monthly quota counters. Requests over the limit get a clean 429 — your upstream API is never hammered by runaway agents.
See every tool call as it happens. Per-tool counts, error rates, and latency in the dashboard. Know exactly how Claude is using your API — before issues reach your users.
Pricing
Start free. Upgrade when you need more projects — calls are always generous.
Try AgentPlay risk-free with your first project.
For teams shipping their first production integration.
Scale across multiple products and teams.
FAQ
MCP (Model Context Protocol) is an open standard that lets AI models call external tools and APIs in a structured, typed way. It was created by Anthropic and is now supported across Claude, ChatGPT, Gemini, Cursor, and a growing list of agent frameworks. AgentPlay hosts an MCP-compliant gateway for your API — one integration that works everywhere.
No. AgentPlay sits in front of your API as a gateway. Your backend doesn't need to know anything about MCP, Claude, or AgentPlay. You just provide your OpenAPI spec and your API key — we handle the rest.
Your upstream API key stays in AgentPlay's encrypted vault — users never see it. Each of your users gets their own AgentPlay API key to connect their Claude account. When Claude calls a tool, AgentPlay authenticates to your API using the shared upstream credential. This makes it ideal for team or internal use where all users share the same access level.
For Claude: open Claude.ai → Settings → Custom connectors → Add new connector, paste the MCP URL, and authorize with an AgentPlay API key. For Cursor and other MCP clients the flow is similar — paste the URL, authenticate, done. The exact steps vary by client but the MCP URL is always the same.
We've got you covered. AgentPlay includes a ready-made AI prompt you can paste into Claude to generate a spec from a plain description of your API. Just describe your endpoints, and Claude will produce a valid OpenAPI 3.0 file you can import directly.
Credentials are AES-256-GCM encrypted before storage. They are only decrypted inside the request handler at runtime, inside the gateway process. They never appear in call logs, user sessions, or response payloads. The key material never leaves the server.
Requests over the limit return a 429 with a clear error message. Your tools stay registered — agents just need to wait or you can upgrade to a higher plan to restore service immediately.
Every week you wait is another week your API isn't inside Claude, ChatGPT, or Gemini. Start on the free plan — your first project is live in under five minutes.
Start for free →Free plan · No credit card required · Cancel anytime
We'll reply within one business day.