Integrations

Gleo MCP

Connect Gleo to your AI assistants — Cursor, Claude Desktop, ChatGPT, and more — via the Model Context Protocol.

MCP — the Model Context Protocol — is a standard way for AI assistants to read from and act on the tools you use. The Gleo MCP server lets you point Cursor, Claude Desktop, ChatGPT, or any other MCP-compatible client at your Gleo workspace and let the AI work alongside you on real playbooks, sessions, and prep.

If you've ever wished you could ask your AI to "draft a kick-off session for this client" or "summarise what came out of last week's retro" without copy-pasting context back and forth, this is the integration that makes that possible.

Video walkthrough — coming soon

What MCP unlocks

Read

Your AI can list playbooks, open templates, read session prep responses, and inspect the agenda for any session you can see.

Draft

Ask your AI to draft a Session Template, propose pre-work questions, or generate an agenda from a goal — all written directly into Studio.

Summarise

Pull a synthesis of pre-work responses, summarise a series of sessions, or extract decisions from a Playbook's history.

Act

Create tasks, add artefacts, and update status from inside your AI client — no context-switching back to Gleo for routine moves.

How it works

The Gleo MCP server lives at a stable URL that your AI client connects to using a personal MCP token generated in Gleo. The token authenticates the client as you, and scopes what it can do to your permissions in your workspace.

You manage tokens — list, create, regenerate, delete — from the same MCP page. Each token shows a short preview of its identifier so you can tell tokens apart at a glance, but the full token value is shown only once at creation. Copy it into your AI client immediately.

Walkthrough

  1. 1

    Open the MCP setup page ~30 sec

    Sidebar → Integrations → Gleo MCP → Set up. The integrations grid is replaced by a dedicated MCP setup sub-page; use the back action to return to the grid.

  2. 2

    Create a token ~30 sec

    Click Create token. Give it a label that reminds you which client it's for ("Cursor — laptop", "Claude Desktop"). Copy the token value before closing the modal — it won't be shown again.

  3. 3

    Install in your AI client ~2 min

    The MCP page has install-instruction tabs for Cursor, Claude Desktop, and Other. Pick yours, copy the config snippet (with the token already filled in), and paste it into your client's MCP config file. Restart the client.

  4. 4

    Confirm the connection

    In your AI client, ask it something Gleo-flavoured: "list my playbooks" or "what's on the agenda for my next session?" If it answers from your workspace, the MCP is connected.

Install snippets

The setup page generates these for you with your token already in place. The shapes look like this:

Cursor

Add the snippet from the Cursor tab to .cursor/mcp.json in your home directory. The structure follows Cursor's MCP config conventions; restart Cursor to pick up the change.

Claude Desktop

Add the snippet to your Claude Desktop config (claudedesktopconfig.json). Restart Claude Desktop to pick up the change. The same token can power both clients — but a separate, labelled token per client makes revocation much easier.

Other

For any MCP-compatible client, the Other tab gives you the raw server URL and the auth header to use. Most clients accept these via their MCP settings UI without needing to edit a config file by hand.

What MCP can do today

The Gleo MCP server registers 50 tools across the playbooks domain — covering reads, template authoring in Studio, live-playbook authoring at runtime, questions and tasks, and the artefact library.

Coverage is reviewed each quarter and grows with the product. Last reviewed: May 2026 (50 tools). The shape of what's possible is summarised below; the per-tool list is the authoritative reference.

The hierarchy MCP works across

MCP mirrors the same Template-vs-Playbook model the rest of Gleo uses. Templates are blueprints (full read + author); Playbooks are running instances (full read; structural authoring is partial — see current limitations below).

flowchart TD PT["Playbook Template
(blueprint)"] PHT["Phase templates"] STT["Session templates"] AGT["Agenda items"] PRT["Prep / postwork items"] INST{{instantiateplaybooktemplate}} PB["Playbook
(running)"] PHL["Phases
(read-only)"] SES["Sessions"] AGL["Agenda items"] PRL["Prep activities"] QTA["Questions, Tasks, Artefacts"] PT --> PHT --> STT STT --> AGT STT --> PRT PT --> INST --> PB PB --> PHL --> SES SES --> AGL SES --> PRL PB --> QTA

Tools for designing templates

Read and author every level of a Playbook Template — metadata, phases, session templates, and the agenda / prep content inside session templates.

Tool Purpose
listplaybooktemplates List templates accessible to you, with filters.
getplaybooktemplate The full template tree — phases, session templates, agenda, prep, postwork, version, theme.
createplaybooktemplate Bootstrap a new draft template (auto-slug; status forced to draft).
updateplaybooktemplate Update metadata — title, description, slug, what-we-achieve, participant roles, visibility, theme. Owner only.
deleteplaybooktemplate Delete a template. Live playbooks instantiated from it are preserved as snapshots.
publishplaybooktemplate Bump the version number; mark the new version as current. Owner only.
validateplaybooktemplate Pre-flight check for errors and warnings before publish — variable references, structural hints, divider items.
instantiateplaybooktemplate Instantiate a published version into a live playbook.
createphasetemplate Add a phase under a template.
updatephasetemplate Rename, reorder (via order), edit description / body.
deletephasetemplate Delete a phase template (and the session templates inside it).
createsessiontemplate Create a session template under a phase, with full metadata.
updatesessiontemplate Update structural metadata — purpose, scheduling, modality, frequency, roles, constraints, variables.
deletesessiontemplate Delete a session template. Live snapshots are preserved.
createsessiontemplateagendaitems Append typed agenda items (label, break, activity) atomically.
updatesessiontemplateagendaitem Update a single agenda item by id.
deletesessiontemplateagendaitem Remove an agenda item; remaining items renumber to a contiguous order.
createsessiontemplateprepactivities Append typed prep items (text, question, video, divider, file, opinion_scale, contact) to prework or postwork.
updatesessiontemplateprepactivity Update a single prep item.
deletesessiontemplateprepactivity Remove a prep item; remaining items renumber.

Tools for running playbooks

Read everything; author session content, questions, tasks, and artefacts. Structural authoring on running playbooks is partial — see current limitations.

Tool Purpose
list_playbooks List playbooks (id, title, description, status).
get_playbook The full playbook — phases, sessions, template ref.
list_sessions List sessions across the workspace.
get_session Full session details (playbook context, owner).
listsessionprework_responses Privileged read of prep responses. Role-gated; PII opt-in. See Privileged reads.
listplaybookquestions List questions with filters (source, priority, answered) and sort.
listplaybooktasks List tasks with filters (status, assignee, due date) and sort.
listplaybookquestion_tags List tags available on questions (shared in use + caller's own).
listplaybookartifacts List artefacts in the playbook's collection.
create_session Create a standalone draft session attached to a company (title-level only).
createplaybooksession Create a draft session in a playbook phase (title-level only).
createagendaitem / createagendaitems Add agenda items, single or in batch.
updateagendaitem Update an agenda item (reorder, edit content).
deleteagendaitem Delete an agenda item.
createprepactivity / createprepactivities Add prep activities, single or in batch.
createplaybookquestion Create a question (typically AI-derived from chat).
addplaybookquestion_answer Add an answer to a question.
toggleplaybookquestion_reaction Add or remove an emoji reaction on an answer.
attachquestiontag / detachquestiontag Manage question tags (shared visibility; idempotent).
createplaybooktask Create a task (typically AI-derived from chat).

Tools for the artefact library

Artefacts can be scoped to a company collection or to a specific playbook.

Tool Purpose
get_artifact A single artefact, optionally with its body content.
search_artifacts Structured + keyword search.
semanticsearchartifacts AI-powered semantic search (requires an org feature flag).
create_artifact Create a structured artefact in a company collection.
linkartifactto_source Attach a pointer to an external source-of-truth.
createplaybookartifact Create an artefact (embedded or external) inside a playbook collection.

End-to-end workflows MCP supports

The two flows below are useful as starter prompts when you're showing a teammate (or an AI client) what's possible.

Authoring a template end-to-end (Studio). Bootstrap → add structure → refine metadata → populate content → pre-flight → publish → smoke-test.

  1. 1

    Bootstrap

    createplaybooktemplate

  2. 2

    Add structure

    createphasetemplate × N, then createsessiontemplate × M.

  3. 3

    Refine metadata

    updateplaybooktemplate, updatephasetemplate, updatesessiontemplate.

  4. 4

    Populate content

    createsessiontemplateagendaitems, createsessiontemplateprepactivities (prework + postwork).

  5. 5

    Pre-flight

    validateplaybooktemplate — surface errors and warnings before you publish.

  6. 6

    Publish

    publishplaybooktemplate bumps the version and marks it current.

  7. 7

    Smoke-test

    instantiateplaybooktemplate to confirm the published version instantiates cleanly.

Running a live engagement. Discover → inspect → instantiate → read → author → tag and react.

  1. 1

    Discover and inspect templates

    listplaybooktemplates → getplaybooktemplate.

  2. 2

    Spin up an instance

    instantiateplaybooktemplate.

  3. 3

    Read the running playbook

    listplaybooks, getplaybook, getsession, listsessions, listplaybookquestions, listplaybooktasks, listplaybookartifacts.

  4. 4

    Author content

    createagendaitems, createprepactivities, createplaybookquestion, createplaybooktask, createplaybookartifact.

  5. 5

    Tag, react, answer

    attachquestiontag, detachquestiontag, toggleplaybookquestionreaction, addplaybookquestionanswer.

Privileged reads

Most reads are structural and don't expose participant-supplied content. The exception worth calling out:

listsessionprework_responses returns participant responses to a session's prework tasks. It's gated at the MCP boundary on top of the standard tenancy check.

Caller What they see
Session owner All members' responses (treated as facilitator).
Facilitator member All members' responses.
Participant or Observer member Only their own responses; any member_id filter is silently dropped and a notice is returned.
Non-member Authorisation error.

Respondent identity is opt-in via a member_pii parameter:

Level Payload Required role
none (default) Session-member id only Any caller.
identity Id, first name, role Any caller.
full Identity plus email Facilitator only.

The member_pii flag controls only the respondent identity payload — the response body itself can contain user-supplied text or uploads regardless of the flag, because Contact, Question, and File-Upload widgets all accept free input. Treat response content as potentially sensitive in any case.

The tool is rate-limited at 30 requests / minute / user (versus 100 for get_session), and standard MCP audit logging captures every invocation.

Security & governance summary

  • Tenancy. Every cross-company call enforces company isolation.
  • Ownership gates. Template-authoring tools (update, delete, publishplaybooktemplate) require the caller to own the parent template.
  • No PII by default. Structural reads exclude member, owner, and email details. Privileged reads are role-gated and opt-in (above).
  • Truncation. Titles and descriptions are truncated for token efficiency on list-style outputs (255 / 500 chars).
  • Per-tool governance. Every call is rate-limited, complexity-checked, audit-logged, and protected by a per-tool kill switch — so a misbehaving tool can be disabled at the server without taking down the rest of the MCP surface.

Current limitations

The areas where MCP coverage isn't yet full — useful for setting expectations on what an AI agent can and can't do today.

Area Limitation
Live phases Phases on a running playbook are read-only. You can't add, rename, reorder, or delete phases via MCP — those edits live with the template, then re-instantiate.
Live session metadata createsession and createplaybook_session set the title only. Purpose, roles, scheduling, modality, and duration on the live side aren't yet settable from MCP.
Live prep update / delete Live agenda items have full CRUD; live prep activities are create-only. Update and delete on prep activities are template-side only today.
Bulk operations Batch creates exist for agenda items and prep activities but aren't yet extended to phase templates, session templates, questions, or artefacts.
Compile a full template There's no single-call "create a template with all its phases, sessions, agenda, and prep" tool yet — multi-call composition is the current pattern.
Template-level tag seeding Question tags can only be created against a running playbook, not against a template before its first instance.

These gaps are tracked and worked on continuously; they're listed here so AI agents and human operators alike know to fall back to the Studio UI for those cases.

Token hygiene

Treat MCP tokens like passwords. A token grants Gleo access to whichever AI assistant holds it. If a token is exposed — committed to a repo, shared in a screenshot, lost on a stolen laptop — regenerate or delete it immediately from the MCP page.

A few habits that make this easy:

  • One token per client / device. Cursor on your laptop and Cursor on your desktop should use separate tokens, labelled clearly.
  • Regenerate, don't share. If a teammate needs MCP access, they should create their own token from their own Gleo account. Sharing a token shares your identity and your permissions.
  • Audit periodically. The token list shows label and last-used time. If a token hasn't been used in months, delete it.

Disconnecting

The Gleo MCP integration doesn't need a single "disconnect" — instead, you delete the tokens that connect specific clients to it. Delete a token and that client immediately stops being able to talk to your Gleo workspace.

Tips & pitfalls

Start with read-only commands. Have your AI list playbooks, open a template, summarise prep responses. Once you trust the loop, you can move on to draft / create / update commands.

The MCP respects your role. Anything you couldn't do in Gleo yourself, your AI can't do through MCP either — asking it to doesn't change the answer. See Roles overview.

Don't paste tokens into chat. Even with your AI assistant. Tokens belong in config files, not in conversation history.

Related pages

Setup Integrations overview Adjacent The command palette (⌘K) Reference Roles overview