An API is what your app can do. MCP lets AI use it.

Give your AI
the right tools.

Turn an API into tools for Claude, Cursor, and other AI assistants. Test what they can do, set the boundaries, and see every call.

Works with any MCP client, includingClaude ClaudeCursor Cursor Visual Studio Code OpenAIWindsurf Windsurf
Try it herePaste → Test → Connect
1
Try
21 AI tool from cURL

Your key stays out of the tool. You add it once as a secret when you deploy.

7

checks on every call, before anything runs

1,000

free sandbox calls. No card, no signup

4

input formats: cURL, OpenAPI, Swagger and Postman

Live demo

Connect your agent.
See it work in 30 seconds.

Get a real MCP URL in one click. Add it to your agent, ask it something, and watch each call arrive here as Accord checks it.

  1. Create a sandbox URLOne click. No account needed.
  2. Add it to your agentClaude Code, Cursor, VS Code or the Claude app.
  3. Ask it to do somethingEvery call shows up live, with the decision and the time it took.

Pick what your agent should do.

Role-based access. Tools that touch production are filtered out of the agent’s tool list.

No signup. Free for 1,000 calls. Responses are sample data.

What happens on every call

A firewall for AI agents.
Seven checks. Every call.

Explore the sandbox’s seven checks, inspect every decision, and see exactly what comes back.

Your agentAccord checksYour APIAccord trimsYour agent

Stops runaway loops

If an agent sends the same call five times in a minute, Accord stops it before it runs up your bill.

Try it: send the same message six times in the sandbox chat.

Each agent sees only its tools

Give a support agent read-only tools. Keep production tools out of its list entirely.

Role-based tool lists, matched by tool name or path.

Risky actions stop for review

Configured risky tools are blocked before execution.

Approval and resume workflows are planned.

Smaller responses, fewer tokens

Links, metadata and fields you do not need are removed before the response reaches the model.

Every receipt shows the real bytes saved on that call.

Private fields never reach the model

Emails, phone numbers or any field you name are replaced first. If that step fails, the response is withheld.

This check fails closed, by design.

Repeat questions answered from cache

Identical read calls come back from the cache, so your API is not hit twice for the same answer.

Exact-match cache, per gateway.

Use cases

Real work for your agents.
With the guardrails built in.

PagerDuty DevOps and IT

Let the agent fix staging. Keep it out of production.

Your agent reads PagerDuty incidents and restarts staging servers on its own. The production restart tool is hidden from it, so it cannot even ask.

Role-based access. Tools that touch production are filtered out of the agent’s tool list.
Read the use case
ZendeskStripeCustomer success

Refund drafts without handing over customer data.

Your agent reads the Zendesk ticket and the Stripe charge, then drafts the refund. Emails and phone numbers are removed before the model sees them.

Private fields removed. Emails, phones and addresses are replaced before the response reaches the model.
Read the use case
HR and onboarding

New hire accounts, approved by a person.

Your agent prepares the Google Workspace account and the Slack invite. Nothing is created until an admin approves it.

Human approval. Every account change is held until a person approves it. Nothing is sent before that.
Read the use case
SnowflakeData and BI

Ask the same question twice. Pay for it once.

Your agent writes SQL against Snowflake. When it runs the same query again, Accord answers from its cache instead of your warehouse.

Cached repeat queries. Identical queries within five minutes are served from the cache.
Read the use case

Deploy

Ready for your real API?
Deploy your own server.

Your converted tools can run as your own MCP server. Your API key stays on the server, so the agent never sees it.

Vercel

One click. Add your gateway’s manifest URL, your API key and a token for your agent.

Deploy to Vercel

Railway

Deploy the same template from GitHub. It takes about two minutes.

Railway steps

Docker

Run it anywhere with one command. You need Docker and a gateway id.

curl -sL https://mcp.edenbuilds.me/docker | bash -s -- YOUR_GATEWAY_ID
Don’t want to manage servers?

Try a hosted sandbox with sample responses. Managed real API hosting is on the roadmap.

Get a gateway URL

How Accord compares

MCP is the socket.
Accord is the circuit breaker.

API gateways were built for apps calling APIs. Automation tools run fixed recipes. Accord is built for agents that decide what to do next, and need limits while they do.

See the full comparison

Pricing

Free while you try it.
Fair when you ship it.

The sandbox is free today. The paid plans below are proposed and open after the private beta. We take no payment now.

Sandbox

$0

Available now

  • Converter and sandbox chat
  • Live MCP gateway URL
  • 1,000 calls for 30 days
  • All seven checks
  • No signup
Create a sandbox

Pro

$79

Proposed, per month

  • Your real API behind the gateway
  • 100,000 calls a month
  • 7-day audit log
  • Custom gateway domain
Join the waitlist

Enterprise

Let’s talk

Planned, annual

  • SSO and identity mapping
  • Slack and Teams approvals
  • Private data plane
  • Log export to Datadog or Splunk
Join the waitlist

Questions

Straight answers
for CTOs and SREs.

What does Accord do, in plain words?
It turns an API into tools an AI agent can call, then sits between the agent and the API. Every call is checked: is this agent allowed to use this tool, is it stuck in a loop, does a person need to approve it, and which fields should the model never see.

Your agents, with limits
you set yourself.

Start with one cURL command. Ship with every call checked.