The MCP Gateway for Developers

Turn Any MCP Server Into an
OpenAI-Compatible API.

Point the gateway at a local or remote MCP server. It auto-discovers your tools, generates OpenAI function schemas, and routes prompts — with streaming, JSON mode, and function calling out of the box.

Start Free See How It Works →
Works with
Cursor Continue Zed LM Studio Devin Any OpenAI SDK + more
3+ OpenAI endpoints supported
Minutes to connect an MCP server
Zero hand-written function schemas
100% OpenAI SDK compatibility

Stop wrapping tools. Start shipping agents.

Your MCP tools deserve a real API. OpenTool API removes all the glue code between your prompt and your tools.

Plug-and-Play MCP

Point at any MCP server — local or remote. Tools are auto-discovered, schemas auto-generated, prompts auto-routed. That's the magic moment.

Drop-In OpenAI Compatibility

Swap the base URL and you're done. Chat completions, models, responses, streaming, function calling, and JSON mode — every OpenAI SDK just works.

Intelligent Tool Routing

The gateway acts like a supercharged OpenAI model that decides when to call tools — LLM-based selection with rule-based overrides you control.

Live in minutes, not weeks

From an MCP server to a production-ready OpenAI endpoint in four steps.

1

Point at Your MCP Server

STDIO or HTTP transport — a local script or a remote host. The gateway spawns, monitors, and restarts it for you automatically.

2

Tools Discovered Automatically

Every tool is discovered and normalized into clean OpenAI function schemas — provider quirks translated for you, no manual mapping.

3

Call It Like OpenAI

Point any OpenAI SDK at your gateway URL. Streaming, function calls, and JSON mode just work — in Cursor, Zed, Continue, or your own code.

4

Monitor Everything

Real-time logs, tool-call traces, latency metrics, error inspection, and MCP server health from one clean dashboard.

Everything you need. Nothing you don't.

Built for developers by developers who understand how MCP servers and OpenAI SDKs actually work.

Full OpenAI API Surface

/v1/chat/completions, /v1/models, /v1/responses — plus function calling and JSON mode, drop-in compatible.

Real-Time Streaming

Token-by-token text/event-stream responses, exactly like OpenAI's — so chat UIs and IDEs feel instant.

Developer Dashboard

Logs, tool-call traces, latency metrics, error inspection, API keys, and usage analytics — all in one place.

Hosted or Self-Run

Use the hosted cloud, bring your own MCP server, or run everything locally with the CLI: opentool serve --mcp ./server.py.

Error Handling & Health

If an MCP server crashes or a tool call fails, the gateway restarts, retries, and alerts you — with a full audit trail of every call.

Enterprise Security

SSO (Google, Microsoft, Okta), team API keys, role-based access, audit logs, private tool hosting, and VPC deployment.

Developers are shipping faster with OpenTool API

★★★★★

"I pointed it at my MCP server and had a working OpenAI endpoint in under five minutes. No schema writing, no glue code. It just worked in Cursor."

I
Indie Developer Building AI-powered dev tools
★★★★★

"The tool routing is what sold us. It feels like a supercharged OpenAI model that knows when to reach for our internal tools — not just a wrapper."

P
Platform Engineer Series B AI startup
★★★★★

"We rolled it out to the whole team with SSO and shared MCP servers. The audit logs and tool-call traces made our compliance review painless."

C
CTO Enterprise SaaS company

Pricing that scales with you

Start free. Upgrade when your tools go to production.

Free
$0/mo
  • 1 MCP server
  • 1 API key
  • 100 requests/day
  • No streaming
Most Popular
Pro
$29/mo
  • Unlimited MCP servers
  • Unlimited API keys
  • Streaming responses
  • Dashboard & logs
  • Tool routing rules

Frequently asked questions

Yes — /v1/chat/completions, /v1/models, and /v1/responses, with streaming, function calling, and JSON mode. Any OpenAI SDK or tool (Cursor, Continue, Zed, LM Studio, Devin) works by just changing the base URL.

Both STDIO and HTTP. The gateway spawns and manages local STDIO servers for you — including monitoring and automatic restarts — or connects to remote HTTP servers.

No. Tools are auto-discovered from your MCP server and normalized into clean OpenAI function schemas automatically — including translating provider-specific quirks.

Yes — a downloadable CLI runs the whole gateway on your machine: opentool serve --mcp ./server.py --port 8080. Or use the hosted cloud and bring your own MCP server.

The gateway behaves like an OpenAI model that decides when to call tools — LLM-based selection combined with rule-based routing and developer-defined overrides. Every decision is visible in the tool-call traces.

Ship tool-calling apps this afternoon.

Connect an MCP server, get an OpenAI-compatible endpoint, and let the gateway do the reasoning.

Start Free

Free tier available. No credit card required.