NewManage your workspace from chat

All your tools through one MCP.

Connect any in-house or public API or MCP server, use up to 99.8% fewer tokens, and manage everyone’s access in one place.

mcpapi · 3 tools

Works with any MCP client

  • Claude
  • ChatGPT
  • Claude Code
  • Codex
  • Cursor
  • VS Code
  • GitHub Copilot
  • Windsurf
  • Gemini CLI
  • Zed
  • Cline
  • Le Chat
  • Raycast
  • n8n

Why teams use MCPAPI

Fewer tokens. Fewer hours. No surprises.

Your agents get every API without the context bill, and your team gets one place to run them.

Tour the product
Spend fewer tokens, get better answers
Connected directly, Stripe’s 612 operations put about 184,000 tokens of tool definitions in front of your agent, sent with every request: input you pay for, or limits you hit sooner. Through MCPAPI it’s about 400, whatever you connect, and the agent keeps its context for your work.
Save hours on every connection
Every API, MCP server and key in one dashboard. MCP servers are checked every night for new tools; an API re-imports from its spec in one click. Set it up once and every AI tool your team uses has it.
Know what every agent can do
Owners, admins and members, read or write per API for each person, and keys narrowed to what a script needs. Whoever runs your AI systems sees every connection, agent and call in one place.

Stripe’s 612 operations, in 400 tokens.

Connected directly, every definition is sent with every request: input you pay for on each turn, and room your agent can’t think in. Through MCPAPI it carries three tools and looks up the one it needs.

Stripe’s operations as tools: about 183,600 tokens. Through MCPAPI: about 400.

Get started in minutes

Here’s how it works

From a spec to an agent that uses it, without writing a server or filling its context.

  1. 1 · Add an API

    Before

    Hand-writing an MCP server per API, or loading one that lists every endpoint as a tool.

    Paste a spec URL

    An OpenAPI or Swagger spec, a Postman collection or an MCP server’s URL. We read the name, address and sign-in from the spec and index every operation. Or enable a public connector.

    Browse public connectors
  2. 2 · Connect your agent

    Before

    One server per API in every client’s config, each with its own key pasted in.

    One URL, every API

    mcpapi.one/mcp. Claude, ChatGPT, Cursor and VS Code sign in with your account; Claude Code, Codex and scripts can use a key.

    See the plans
  3. 3 · Ask

    Before

    Hundreds of tool definitions in context before the agent reads your first message.

    Three tools, then the work

    Your agent searches in plain words, reads only the operation it picked and runs it. Every call is checked against your rules and logged.

    Read the FAQ
MCPAPI’s Overview: the context saved, the connected APIs and MCP servers, and the agents’ recent calls.

The MCPAPI connector

Run your workspace from the chat you already use.

Adding APIs, inviting people, setting access, making keys and reading logs are tools too, found by the same search and run with your role. Billing and approvals stay in the dashboard.

Asking from chat: add GitHub’s API from its spec and give Ana write access to it.
New chat mcpapi · 3 tools

For whoever runs your AI systems

Everything you get, in detail

Save tokens

The right tool, first time
Agents ask in plain words and get five matches with their risk. On our Stripe test queries the right operation is in the top five 96% of the time.
Costs you can predict
One credit per successful run. Searches, tool details and failed runs are free, so an agent finding its way costs you nothing.

Save time

Set up once, use everywhere
One URL for Claude, ChatGPT, Cursor, VS Code and any MCP client. Switch AI tools without setting anything up again.
No servers to build or host
Most APIs have no MCP server. Paste the spec and it’s ready; we host it, index it and keep it running.
Problems come with their fix
An expired key or a refused sign-in comes back with a link to the page that fixes it, and the Overview lists problems first.
Saved prompts
Write a task once and every connected client lists it, for you or the whole workspace.

Stay in control

Agents can’t go rogue
Read-only keys never write, and an API can ask the person before every write. An agent can never approve itself.
Your keys will always be safe
Encrypted with AES-256-GCM and used only to make the call. They never reach the model, the chat or a log.
One step to cut access
Remove someone and their keys, grants and app sign-ins go with them.
A full audit trail
Who, which agent, which operation, which account, how long it took and why it failed.
Test without risk
Sandbox and live side by side, and each person can connect their own account.

Public connectors

Ready-made, or bring any API

Enable a public connector and add your own key. Not listed? Any OpenAPI 3.x or Swagger 2.0 spec, one file or many, a Postman collection or a remote MCP server.

Browse all 6 connectors

Pricing

Searching is free. You pay for runs.

A successful run is one credit. Searches, tool details and failed runs cost nothing.

Full pricing

Free

$0 forever

  • 100 credits a month
  • Every public connector
  • The three MCP tools, for any agent
  • Teams, roles and per-API access
Start free

Plan

Launch: 7 days free, then 50% off for 6 months

$19 a month

  • 5,000 credits a month (1,000 during the trial)
  • Up to 5 of your own APIs or MCP servers
  • Reranked search: the best 5 of the top 20
  • Buy more credits whenever you need them
Start 7-day free trial

FAQ

Questions, answered

Anything else? Sign in and ask the MCPAPI connector from your chat.

Ready to get started?

Every API for your agents. One place for your team.

Fewer tokens on every request, every connection in one dashboard, and the access each person needs.