Use Stated with MCP

Connect an AI client or agent to Stated through MCP to create, update, share and retrieve durable human-facing publications.

2026-09-15 · 2 min read

Model Context Protocol (MCP) lets an AI client or agent use Stated as a publishing destination while it is doing the work.

Use MCP when the agent needs to create a Document, Page or Experience, update that same publication later, share it with people or retrieve its results. The publication becomes the durable human-facing output of the workflow.

Connect and authenticate

Connect the Stated MCP server from the AI client or agent environment you use. Authentication identifies the Stated account that owns the resulting publications.

Keep credentials and access tokens in the agent’s secure configuration. Do not place them in prompts, source documents or public Pages.

Design around one durable publication

The default pattern is:

1. Resolve whether a publication already exists for the business object.
2. Create it only when it does not.
3. Store its Stated identifier or URL in the workflow state.
4. Update that same publication on later runs.
5. Share or retrieve results only when the workflow needs them.

This prevents a recurring agent from producing a new disconnected report every time it runs.

Ask for outcomes, not low-level formatting

An agent instruction should state the audience, intended action and required evidence. For example:

Create or update the Stated Page for this account. Lead with the current recommendation, retain the evidence and caveats needed to support it, ask one decision question, and keep the existing Page link and access setting.

Let Stated handle the publication format. The agent should not try to recreate the Page as raw HTML or a copied chat transcript.

Handle access and limits

Some capabilities depend on the owner’s plan, available credits or connected provider. A successful tool call is the only confirmation that a capability was used.

When a request is denied, handle the returned reason explicitly. Common outcomes include missing authorization, a plan requirement, insufficient credits, a publication limit or a temporarily unavailable provider. Do not retry paid or provider-backed actions blindly.

Keep human control where it matters

An agent can create, update and prepare work, but it should not assume it can make consequential decisions for a person. Use a Stated interaction, approval step or human review before an external action when the decision requires accountability.

Next steps

See Agentic Publishing with Stated for the overall architecture, Persistent AI Artifacts and Idempotent Workflows for safe retries, and Build Workflow Deliverables with the Stated API for API-oriented workflows.