Outlign has a built-in MCP (Model Context Protocol) server that lets AI assistants like Claude, Claude Code, and Cursor, securely connect to your Outlign workspace. Once connected, your AI assistant can look up projects, create and update tasks, post messages, add comments, and more, all through normal conversation.
The AI assistant acts as you: it can only see and do what your own Outlign account can see and do. Internal (team-only) content stays internal, and client users only ever see client-facing content.
Find things: search across projects and tasks, list projects, clients, templates, phases, tasks, messages, and milestones
Manage projects: create projects (from scratch or from one of your project templates), update project details
Manage tasks: create tasks in a specific phase, update them (complete, due dates, content), assign team members
Communicate: post messages to a project's client feed or team-only feed, add and edit comments on tasks
Plan: create and update phases and milestones
All content is exchanged as plain Markdown — your assistant reads and writes the same rich text you see in Outlign.
You don't need an API key — connecting works like "Sign in with Outlign":
In your AI tool, add a new MCP server (sometimes called a "connector" or "integration") with this URL:https://go.outlign.co/mcp
Claude (web/desktop): Settings → Connectors → Add custom connector
Claude Code: claude mcp add outlign --transport http https://go.outlign.co/mcp
Cursor / VS Code: add the URL in the MCP settings for your editor
Your browser opens an Outlign login page. Sign in with your normal Outlign account (if you're not already signed in).
Review and approve the access request.
That's it — your assistant is connected and can start working with your Outlign data.
Access tokens are short-lived and refresh automatically for up to 90 days, after which you'll simply be asked to sign in again.
Morning catch-up: "What's on my plate today in Outlign? Anything overdue?"
Meeting notes → tasks: "Here are my notes from the ACME kickoff call. Create tasks in the internal Concept phase of the ACME Rebrand project for each action item, and assign the design ones to Sarah."
Project spin-up: "Create a new project for ACME using our Website Design template, and rename the client phases to match their brand launch."
Status updates: "Draft a client-facing status message for the ACME project summarising what was completed this week, and post it when I approve."
Triage: "List all incomplete tasks in the Bugs phase of the Product project and summarise which ones mention Safari."
Handovers: "Reassign all of Tom's open tasks in the ACME project to Priya and add a comment on each explaining the handover."
Tip on internal vs client-facing: every project has team-only (internal) and client-facing phases and message feeds, which often have similar names. A well-behaved assistant will ask which side you mean when it's ambiguous — being explicit ("post this to the client feed", "add it to the internal Concept phase") gets the best results.
You authorise it, you control it. The connection is approved by you in the browser; no passwords or API keys are ever shared with the AI tool.
It acts as you. The assistant has exactly your permissions — same projects, same companies, same internal/client visibility. It can never see anything you couldn't see yourself.
Scoped access. Connections use granular scopes (e.g. read projects, write tasks) rather than blanket access.
Every action is logged. All tool calls made by connected agents are audit-logged.
Disconnect any time. Go to Integrations → Connected AI Agents in Outlign to see everything connected to your account and disconnect it instantly.
"Unauthorized" or connection expired: disconnect the server in your AI tool and reconnect — you'll be asked to sign in again.
The assistant can't find a project: check that your Outlign account actually has access to it, and try giving the exact project title.
Rate limits: heavy automated use may be throttled briefly; the assistant will usually retry on its own.
