How to connect Claude Code, Cursor or Codex to your product plan

Connect the agent to your planning tool over MCP, the open protocol coding agents use to reach tools outside the editor. With Fulveo you add one address to Claude Code, Cursor or Codex, approve which workspaces and permissions the agent gets, and it can read your plan, update tasks and hand finished work back for review. It acts as you, never with more access than you have.

What is MCP, and why does it matter for product work?

MCP, the Model Context Protocol, is an open standard that lets an AI agent use tools beyond the one it runs in. A coding agent with MCP can read from and write to other systems, with your permission.

For product work, that changes where the brief lives. Without it, the plan sits in one tool and the agent in another, and you carry each task across by hand. With it, the agent reads the task where you wrote it, including the context and acceptance criteria, and reports back in the same place.

What can a coding agent do with your Fulveo plan?

With the permissions you grant, an agent can:

• Read tasks and plans, including descriptions, subtasks and acceptance criteria.
• Change tasks: update them, comment on them and move them between columns.
• Read boards, so it sees the flows and decisions behind a task.
• Draw on boards, if you tick it. This one is off by default.
• Read connected code, when the workspace has a GitHub repository.

Everything it does is recorded under its own name, so you can always tell what a person did from what an agent did.

How do you connect Claude Code, Cursor or Codex?

The server address is https://mcp.fulveo.com/mcp.

In Claude Code, run the command below, then run /mcp inside Claude Code, pick Fulveo and sign in:

claude mcp add --transport http fulveo https://mcp.fulveo.com/mcp

In Codex, add the server and sign in:

codex mcp add fulveo --url https://mcp.fulveo.com/mcp
codex mcp login fulveo

In Cursor, choose Add to Cursor in Fulveo's settings, or add the address to ~/.cursor/mcp.json.

The AI agents page in the docs has the steps for Claude on the web or desktop, VS Code, Gemini CLI and Windsurf.

What does the agent get access to?

When you connect, a Fulveo tab opens and asks you to choose two things:

• The workspaces it may use, from the ones you belong to.
• What it may do: read tasks and plans, change tasks, read boards, draw on boards, and read connected code.

The agent never gets more than your own role allows. In a workspace where you can only view, it can only read. On a board it can only delete or restyle what it added itself, so what you and your team made stays out of its reach. You can change its access or disconnect it at any time from Fulveo's settings.

How do you keep a person in charge of what ships?

Give the work a review step the agent cannot skip. In Fulveo's planner, any column can be marked so that only people can move cards into it. The AI-first plan template is set up this way from the start: Backlog, In progress, Human review and Done, with Done reserved for people.

The agent picks up a task, does the work and moves the card to Human review. You check it against the acceptance criteria and move it to Done. The agent can finish work. Only a person decides it is finished.

What if my agent does not support MCP?

Use Copy as prompt on the task. It writes the task, its description, the files it touches and its acceptance criteria into a brief you can paste into any agent. See how to brief an AI coding agent for what makes that brief work.

Read next

• AI agents in the docs: every client, step by step.
• How to brief an AI coding agent: what a good brief contains.
• Pricing: the MCP server is included in Solo and Team.