Every developer we talk to has some kind of AI coding agent open in another window: Claude, Copilot, Codex, whatever the team has standardized on. A genuine teammate, this agent writes code, reviews pull requests, retrieves information, or debugs a stack trace at 2 a.m.
When a conversation turns to the actual site, things get harder: is the build green, did that deploy actually succeed, what changed in the last release? Suddenly, the agent goes quiet. It can see your code, but it can't see your platform.
Pantheon is closing that gap today. We are excited to share that the Pantheon MCP Server is now in Public Beta, meaning the AI tools your team already uses can connect directly to your Pantheon fleet and act on it, in plain language, under your own permissions.
What Pantheon MCP Unlocks
MCP (Model Context Protocol) provides a standard way to connect AI agents to the systems they need to work with. Instead of an agent that only knows your codebase, you get one that knows your codebase and your infrastructure: your sites, environments, deploys, and logs.
With the Pantheon MCP Server, you can ask your agent things like:
- Which of my sites have pending Upstream updates?
- Why did the last build on this environment fail?
- Spin up Multidev for this branch.
- Pull the runtime logs from staging and tell me what errors are showing up.
You don’t have to switch to a dashboard tab or copy-paste build logs into a chat window. The same fleet intelligence that lives in your Pantheon dashboard is now available wherever your agents already live.
Built on the Access You Already Have
How to connect?
Connecting the Pantheon MCP Server takes just a few steps:
- Add the Pantheon MCP Server to a compatible AI client, such as Claude Desktop, Claude Code, or Cursor.
- Sign in with your existing Pantheon account.
- Review the access being requested and approve the connection.
Once connected, your agent can work with Pantheon workspaces, sites, environments, and deployments your account can already access.
Who can connect?
Any Pantheon customer can connect to the MCP Server. There is no waitlist or special enrollment process.
Each teammate who wants to use MCP can connect it in their own compatible client with their Pantheon account. There is no separate site-by-site setup: the agent works wherever that account has permission.
How to Stay in Control?
The MCP Server uses your existing Pantheon identity and permissions. Your client holds the session, and every request the agent makes runs under the permissions of your Pantheon account. That means the agent can only see and do what your account can already see and do in Pantheon. If you can do it in Pantheon today, your agent can help you do it faster.
Read access is available by default. Any action that changes state (rebuilding an environment, rolling back a deploy, setting a secret, or spinning up a new site) pauses until you explicitly approve it in your AI client.
Check our documentation on how to connect the MCP Server in more detail.
What Your Agent Can Do Today
Once connected, your agent has tools covering the parts of the fleet that engineers spend the most time managing:
- Workspaces & sites: list and inspect your workspaces and sites, see who has access, spin up new sites.
- Environments: list and inspect environments, create multidevs on demand.
- Builds & deploys: check build history and status, pull build logs, get help diagnosing a failed build, check deploy and workflow status, sync code, trigger a rebuild, or roll back a deploy.
- Runtime logs: pull runtime logs straight into your conversation.
- Upstreams: view available upstreams and create new ones for a workspace.
- Secrets: view configured secrets and set values, including per-environment overrides.
That's the opening set of tools, and even at this stage it cuts out a lot of the checking-the-dashboard busywork. More is landing when we reach general availability next year, so think of this as a starting point.
Your Fleet, Wherever Your Agents Are
We built the Pantheon MCP Server because our customers were already living in agentic workflows, and the fleet data that used to be locked inside a dashboard needed to go where the agents are. It's part of the same push we've been on all year: making the fleet something your whole team, human or AI, can actually act on. Your feedback helps us build better and faster! Submit your ideas in the Pantheon Slack Community channel here: #beta-mcp-server.
Already a Pantheon customer? Check our documentation on how to connect the MCP Server today and start asking your agent about your fleet.
Still evaluating Pantheon? Talk to us about what running an agent-ready fleet looks like.
Missing any functions or bugs? Join the Pantheon Slack Community channel #beta-mcp-server and post your feedback.
