cypress.DOCS
Log inRegister
Cypress Docs

Apps & Agents

Connect an AI agent or an app to Cypress. The MCP servers, a sandbox to start in, what an agent may and may not do, keys and tokens, and every action of the app.

TL;DRAI summary

This chapter explains how you can connect other software, like AI agents or your own systems, to Cypress. These connections run under the same rules as the main app, so you stay in control of your fleet.

Dive deeper:

Cypress can be worked by software as well as people: an AI agent that speaks the Model Context Protocol (MCP), a partner's app your organization connects, or your own systems. Whatever the door, the rules are the app's. Every call runs the same code as the button on the app's screen, with the same checks, the same floor and the same privacy. A decision that belongs to a person is never made by software: it comes back with a link for a person to act on.

How it fits together

AI agentClaude, Cursor, your ownApp or systema partner's, or yoursMCP serversOperate, Intelligence, ManualREST API/api/v1The app's own codesame rules, your organizationa person's decision: a link to hand overevery change heard by webhook

Three ways in

Way inForStart with
MCP serversAn AI agent. Cypress Operate works in your organization, Cypress Intelligence reads what Cypress knows, Cypress Manual reads these docs (no key needed)MCP quickstart, then MCP servers
REST APIYour own systems and partners' apps, under /api/v1Developer quickstart
WebhooksHearing what changed, whoever changed it: a person in the app, an agent, an automation, an importWebhooks, and Events and sync to keep a copy in step

The MCP tools and the REST operations are the same actions: each MCP tool runs one REST operation, through the same code. Everything a person does on the app's screens is one of them, listed in The app's actions.

Before you go live

  1. Build in a sandbox. A Cypress Manager makes you one, with a test key that can't reach a real person. See Sandbox.
  2. Choose the credential. An API key for your own systems and agents, Connect Cypress for an app many organizations use or an agent platform. See Keys and tokens.
  3. Give it the least it needs. Scopes decide what it may do; toolsets decide which tools an agent is shown. By default an agent sees the core tools, thirteen of them.
  4. Dry run what you can't take back. Every write takes dryRun, and idempotencyKey makes a retry safe.
  5. Measure it. Run the evaluation tasks in a sandbox before it works in your live organization.

In this section

PageWhat it covers
MCP quickstartA sandbox, a client, and an agent's first ten minutes in Cypress
What an agent can and cannot doThe agent role, area by area, and how a refusal comes back
Keys and tokensEvery credential, who it acts as, scopes and toolsets
MCP serversThe three servers, every tool by area, the protocol, switching a server off
Core toolsThirteen tools for an agent's everyday work, and every other operation through discovery
Domain modelThe records, their ids, links and states, what changes them and what they announce
Events and syncKeeping a copy of your records in step: backfill, every change, reconcile
Connecting fromClaude, ChatGPT, Cursor, Copilot Studio, Agentforce, Palantir: which door and which credential
EvaluationsTasks to measure an agent in a sandbox, each with a good run, what it must never do, and the check
SandboxWhat a sandbox holds, how it behaves, starting over
Connect CypressOAuth for apps: the consent screen, tokens for a person or an agent, Sign in with Cypress
The app's actionsEvery button of the app as a REST operation and an MCP tool, with the webhook it sends

For your team rather than your developers, the platform manual explains the same thing in the app's terms: Apps, agents and MCP and Connect an app or an agent.