Your agents, inside ChatGPT and Claude
Cronloop runs a hosted MCP server, so the assistant you already use can work in your Cronloop account. Create agents, start runs, read back what happened, and keep memory up to date, all from the same conversation.
https://app.cronloop.ai/v1/mcpSet up an agent that answers new Zendesk tickets from our help docs every 5 minutes and escalates anything sensitive to me.
agents_createSupport triage, every 5 minutessecrets_setZendesk token stored, write-onlyruns_startFirst run startedruns_getAnswered 6 tickets, escalated 1
Support triage is live. Its first run answered 6 tickets from your help docs and escalated one refund request to you. It will keep running every 5 minutes.
Connect in about a minute
Nothing to install and no key to manage. The connection is a normal sign-in, and you can remove it from your client whenever you want.
- 1
Add the address
Open the connector settings in Claude, ChatGPT, or any other MCP client and add Cronloop as a custom connector.
- 2
Approve the connection
Sign in to Cronloop and approve access. It is a standard OAuth sign-in, so there is no key to copy, paste, or keep safe.
- 3
Ask for what you need
Your assistant now speaks to Cronloop directly. Ask it in plain language and it does the work in your account.
What the sign-in supports
- OAuth 2.1 sign-in with PKCE, so no key is ever copied or stored in your client.
- Dynamic client registration and Client ID Metadata Documents, so most clients connect with the address alone.
- Refresh tokens keep the connection working without asking you to sign in again.
- An agent API key also works as a bearer token when you want a connection limited to one agent.
The server speaks Streamable HTTP MCP and is stateless, so it works with clients that support remote servers rather than local processes.
What your assistant can do
A small, stable set of tools that covers the whole life of an agent. Everything an owner can do in the dashboard for agents, runs, settings, and memory.
Agents
Create an agent, rewrite its instructions, change its schedule, then pause, resume, or delete it.
agents_listagents_getagents_createagents_updateagents_pauseagents_resumeagents_deleteRuns
Start a run on demand, follow it to the end, cancel it, and review what recent runs actually did.
runs_startruns_listruns_reviewruns_getruns_cancelSettings and secrets
Add the settings and credentials a job needs. Values are encrypted on the way in and can never be read back out.
configuration_listenvironment_setenvironment_deletesecrets_setsecrets_deleteMemory
Read, write, and clear the durable memory each agent carries from one run to the next.
memory_listmemory_readmemory_writememory_deleteThings people ask for
Say it the way you would say it to a teammate. Your assistant picks the tools and reports back with what it did.
“Create an agent that reads new GitHub issues every hour and drafts a reply for each one.”
agents_create
“Why did the release monitor fail last night, and what would you change?”
runs_review
“Pause outbound outreach until Monday morning.”
agents_pause
“Save this Linear API key on the triage agent and start a run.”
secrets_set
Access you can reason about
A connected assistant gets the tools above and nothing else. The limits are enforced on every request, not just in the client.
- A connection acts as you, and reaches only the agents in your own account.
- Saved settings and secrets are write-only. They can be replaced or removed, never read back.
- Billing, account settings, and saved connection credentials stay out of reach.
- Every call is recorded with the tool name and result, without its arguments or contents.
Put an agent to work from your next message.
Create a free Cronloop account, connect your assistant, and let it run the loop for you.
