PrimeFlow and MCP
Let an AI agent read your PrimeFlow canvases, create new ones, and place nodes, connections, and whole maps on your behalf.
PrimeFlow is one of the places where an AI agent can do real work. Connect Claude, an MCP-compatible editor, or any other MCP client to PrimeTask, and your agent can see your canvases, read them, create new ones, and place nodes, connections, and entire maps for you.
A brief you've been staring at becomes a project canvas. A backlog becomes a sprint map. A PRD becomes a goals-and-milestones plan. A call with a prospect becomes a stakeholder diagram. The canvas is no longer something you build alone - it's a surface your agent can shape alongside you.
This article covers what PrimeFlow exposes through MCP and how to put it to work. For MCP setup, transports, tokens, and permissions, see bring-your-own-ai and External Integrations Settings.
You choose what the agent can do
MCP read and write permissions are separately controllable. You decide whether the agent can just look, or also create and edit.
What an agent can do on your canvases
Once MCP is set up, your agent has a rich PrimeFlow toolkit. It can:
- Discover your canvases - list all canvases in the active Space, with their names and structures.
- Read a canvas - every node, its type, its position, the connections between them.
- Create a new canvas and give it a name.
- Rename or delete a canvas.
- Add nodes to a canvas - individually or dozens at once.
- Move and update nodes - change positions, update labels, replace content.
- Connect nodes with labeled connections.
- Remove nodes and connections.
- Build a whole canvas in one atomic step - many nodes, all their connections, in a single request. No half-built maps.
Node types your agent can place
PrimeFlow exposes the full node family to MCP, with the same tier rules the product uses everywhere else.
Available on every PrimeTask license
- Mind map - Central Topic, Main Topic, Sub Topic, Floating Topic
- Work - Project, Project Goal, Project Milestone, Project Note, Task, Checklist
- Annotations - Sticky Note, Text Sticker, Emoji Sticker
- Media and links - YouTube, X, Image, Web Link
- Structure - Group, Quick Notes, Documentation
Pro tier
- File family - File, Folder, PDF, Audio File, Video File, Code File, Text File, Compressed File, and the Image File view
- Navigation - Portal, Cross-Space Portal
- Integrations - Third-party app / deep link
- Diagrams - Mermaid
CRM (Pro with CRM enabled)
- Contact, Company, Activity
If the agent tries to place a node your license doesn't cover, PrimeTask returns a clear message and the action stops. Your canvas is never left in a half-built state.
Use cases - real things agents can do
Turn a PRD into a project brief canvas
Paste a product requirements doc into your agent. Ask it to build a PrimeFlow canvas with a Project node, goal nodes for each outcome, milestone nodes for each phase, and seed Task nodes for the obvious first items - all connected into a plan you can start editing.
Draft a sprint canvas from a backlog
Hand the agent a list of issues - from Linear, GitHub, or a doc - and ask it to create a sprint canvas. It places a Task node for each item, groups them by area, connects blockers with labeled edges, and you open a map that's ready for your sprint planning meeting.
Map stakeholders for a new account
Ask the agent to build a stakeholder canvas for a company in your CRM. It places the Company node, the key Contact nodes around it, and labeled edges showing who reports to whom or who works with whom. Your next kickoff starts with a picture, not a spreadsheet.
Convert a markdown outline into a mind map
Point the agent at a long markdown document and ask for a mind map. It reads the structure, places a Central Topic, Main Topics for each heading, Sub Topics for each sub-heading - and your outline becomes a canvas you can think in.
Build a weekly review canvas
Every Friday, ask the agent to draft a weekly review canvas: what was completed, what's still open, what's coming, what's worth reflecting on. The agent assembles the nodes; you annotate with sticky notes.
Generate a retrospective canvas
After a project, ask the agent to scaffold a retrospective - what went well, what didn't, what to try next, grouped with sticky notes your team can vote on.
Plan a learning canvas for a YouTube series
Give the agent a playlist link and ask for a study canvas. It places a YouTube node for each video with a Documentation node next to it for notes, connected in order. You study the series with a canvas that grows as you do.
Draft a customer journey canvas
Describe the stages of a customer journey in chat. The agent builds a canvas with nodes for each stage, stickies for the customer's actions, and connections between the handoffs.
Kick off a product launch canvas
Ask the agent to sketch a launch canvas from a short description. It drops timeline stickies, milestone nodes, task nodes, CRM contact nodes for the people involved, and connects them into a launch plan.
Write a post-mortem canvas
Give the agent the incident timeline. It creates a canvas with a timeline, nodes for contributing causes, Task nodes for actions to take, and connections that show cause and effect.
Kick off a personal goal canvas
Tell the agent you want to set a three-month goal. It builds a canvas with an outcome node, milestone nodes for checkpoints, Task nodes for habits, and checklist nodes for the first steps.
Start with a conversation, end with a canvas
The best prompts describe the outcome you want - "a sprint canvas for next week" - and let the agent assemble the nodes. Edit afterwards.
Build complex maps in one step
The agent can build a whole canvas in a single atomic step - dozens of nodes, all the connections, all the labels - in one request. Either the whole canvas is built, or none of it is. You never end up with half a map, dangling edges, or partial structures.
This is what makes an agent useful for real work on PrimeFlow. A "build this canvas for me" prompt turns into a complete, connected map, not a stream of individual edits.
Setup in brief
Step 1
Open Settings → External Integrations.
Step 2
Enable the MCP Server.
Step 3
Choose your transport and connect your MCP client with the token PrimeTask gives you.
For the full setup guide, transports, permissions, and troubleshooting, see bring-your-own-ai.
Things worth knowing
Agents work on the active Space
The agent builds on the Space you currently have open. If you want it to work somewhere else, switch Spaces first or ask it to switch for you.
Canvas writes are serialised
If two MCP calls try to modify the same canvas at the same time, PrimeTask runs them in order. The canvas never corrupts from overlapping edits.
CRM nodes require your CRM license
Contact, Company, and Activity nodes require Pro with CRM enabled. The agent will tell you if it can't place a CRM node on your current license.
Sensitive data is never shared
Passwords, tokens, PINs, encryption keys, and similar fields are always removed before any MCP response leaves PrimeTask. What the agent sees is what you see on screen.
Read and Write are separate permissions
Let an agent read your canvases without being able to change them, or grant full access, or anything in between. You're always in control of what the agent can do.
No silent failures
When the agent tries something your license doesn't allow, PrimeTask returns a clear message. You and the agent both know what happened.
Common questions
"Which AI models can use this?"
Any MCP-compatible client - Claude Desktop, Claude Code, MCP-compatible editors, and any other agent that speaks MCP. See bring-your-own-ai for the current list and setup steps.
"Can the agent see every canvas I have?"
The agent sees what you give it access to. When the MCP server is enabled for a Space and Read is turned on, the agent can list and inspect canvases in that Space. Read-only mode lets the agent look without making changes.
"What happens if the agent tries a Pro node on my Standard license?"
PrimeTask returns a clear "this node requires Pro" message and the action stops. Your canvas isn't touched.
"I opened my canvas and there's a new one I wasn't expecting. What happened?"
The agent likely just finished building one for you. Canvases created by an agent appear in the same library as any other canvas. Delete them or rename them like any other canvas.
"Can two agents work on the same canvas at the same time?"
Yes - PrimeTask serialises the writes so each one completes before the next begins. You won't see a corrupted state.
"How do I stop an agent that's running?"
Most MCP clients have a stop action for an in-flight conversation. Use your client's stop action. PrimeTask applies changes as they come in, so stopping mid-way just means no further changes arrive.
Where to go next
| If you want to… | Read this |
|---|---|
| Set up MCP in detail | bring-your-own-ai |
| Configure external integrations | External Integrations Settings |
| Understand the nodes your agent can use | Adding Nodes in PrimeFlow |
| See what's available on each license | PrimeFlow Standard vs Pro |