AI clients and MCP

Connect an AI client to ask for help with Fieldbook content, media, feedback, and learning reports. Model Context Protocol (MCP) gives the client individual tools; your account and approved permissions determine which ones it can use.

Search and AIUpdated Oct 5, 2026v1

Ask an agent to use Fieldbook

Describe your task in a connected client such as ChatGPT or Claude. The AI chooses and calls Fieldbook’s tools; you don’t need to know their names. The client can explain missing permissions or direct you to Fieldbook’s interface. The reference below lists all supported tools; your connection may have fewer.

Access follows your account and permissions

Your Fieldbook role determines available work:

Role	Available work
Learner or guest	No MCP tools
Manager	Managed-team reports
Contributor	Content, media, feedback
Administrator	Content, media, feedback, assignments, all-team reports

Managers need an explicitly managed team; reporting includes its descendants. Contributors with manager responsibility can report on those teams too. Publishing access alone does not grant learner reports. Team and learning-group filters only narrow this scope.

Consent can narrow access, but cannot grant a new role or responsibility. Fieldbook checks permissions on every call. See Roles and permissions.

Consider what the external client can read: unpublished drafts and Course answer keys through content access, comments and respondent names through feedback access, and named learners through reporting access. Ask AI is separate: it answers reader questions from selected published prose inside Fieldbook.

Connect your AI client

The operator must enable MCP first; Set up MCP covers configuration. The browser-local demo has no MCP server. Use a client that supports authenticated remote MCP connections.

Get your installation’s server address. Contributors and Administrators can open Manage content or Manage organization, then MCP, and copy Server address. Managers can ask their Administrator.
In your AI client’s custom MCP connection settings, add that address. The address ends with /api/mcp, for example https://fieldbook.example.com/api/mcp (opens in a new tab).
Follow the client’s sign-in flow using your active, registered Fieldbook account.
On Fieldbook’s consent page, confirm the account and client. Under Approve tool permissions, select the capabilities you intend to use, then Allow connection. Select Deny to cancel.
Return to the client and ask, “What can you help me do in Fieldbook?”

Client screens differ. The client supplies its displayed name; approve a connection you initiated.

Direct the work and check the result

For editing, identify what should change or stay intact. Ask the client to read the latest content, save a draft, and check the result. Review the draft’s wording and placement in Fieldbook’s editor. Have the client reconcile any concurrent edits before saving again.

Saving a draft leaves the published version unchanged. Request publication, unpublication, or assignment changes explicitly. When publishing, specify whether a Course change should require learners to complete a new version, or an Update should return to the top of the feed. Ordinary corrections do neither. After publication, check the reader view. If an action’s result is uncertain, ask the client to check Fieldbook before repeating it.

For reports, confirm the included audience and whether the client retrieved the complete report. Assigned completion and optional learning are separate measures; partial results need to be identified. Progress and reporting explains these measures. For feedback analysis, keep your instructions separate from the comments being analyzed.

For media, ask the client to reuse an uploaded image or video. New uploads require it to transfer the file itself; some clients cannot. If so, upload through Fieldbook’s editor, then ask the client to find it. Use only a verified, ready file in content. A temporary upload address is not a reusable media link, and MCP cannot import an arbitrary remote URL.

Review or revoke a connection

Open Manage connections from Fieldbook’s MCP page, or visit /connections on your installation.

Select Review permissions, adjust capabilities, then Approve selected permissions. Refresh tools in your client and start a new conversation. Newly requested capabilities need consent; becoming an Administrator requires approving permissions again.

Select Revoke access to stop a connection’s Fieldbook tools, or select several connections and choose Revoke selected connections. Reconnecting requires authorization again. If Fieldbook disables access but reports that the identity provider could not revoke saved consent, retry before reconnecting.

Available tools reference

The AI selects tools for your request. Availability depends on your account, consent, and installation. Clients may prefix these names.

Tool	What it lets the client do
`get_capabilities`	Check available actions, missing consent, limits, and manual alternatives.
`search`	Find published content and drafts, including lesson prose.
`fetch`	Read a complete current draft, lessons, and quiz answer keys.
`get_authoring_options`	Find existing Docs destinations, categories, and supported formats.
`create_content`	Create a Doc, Update, or Course draft.
`update_content`	Edit an existing draft, including lessons and quizzes.
`publish_content`	Publish a draft when you request it.
`unpublish_content`	Remove a publication while retaining its draft and history.
`list_media`	Find ready uploaded images and videos.
`prepare_media_upload`	Prepare a new image or video upload.
`complete_media_upload`	Verify a transferred file and obtain its reusable reference.
`get_reporting_scopes`	Identify permitted teams, learning groups, and published courses.
`learning_report`	Read scoped, named progress, assignment, due-date, and overdue reports.
`feedback_report`	Read content or general feedback and respondent names.
`content_report`	Read Course activity and feedback totals; Administrator only.
`manage_course_assignment`	Assign or unassign a course for an audience; Administrator only.

For client implementers, the generated MCP contract (opens in a new tab) defines tool inputs, outputs, and execution rules.