Key Takeaways
Craft.io MCP Server: Your portfolios, workspaces and feedback portals are now accessible to any AI agent via an official, remote-hosted MCP server.
Read and Search Your Product Data: Find any work items or OKRs, and query your backlog directly inside Claude, Cursor, or any other AI tool.
Create Items and Capture Feedback: Enable AI agents to generate PRDs or capture customer requests and create them as a work item or a feedback item.
User-based API Keys With Scoped Access: Every connection to the MCP server is tied to a specific user, with controls over what data and actions each key can access.
Agent Actions Are Tracked: The activity log shows exactly how any change was made - by an AI agent or through the UI - while preserving each user's accountability.
Craft.io MCP Server
AI tools are everywhere in product teams today. But they have been working blind. When a PM asks Claude to help write a PRD, Claude does not know what is already on the roadmap. When an engineer asks Cursor to implement a feature, Cursor has no access to the spec. The tools that are supposed to accelerate product work end up producing generic output, because they have no connection to where the actual product context lives.
Craft.io's MCP fixes this. Your product data - your features, your roadmap, your customer feedback - is now part of the context your AI tools can reason about. Connect your workspace to any MCP-compatible tool - like Claude Desktop, Cursor, or VS Code - to start leveraging your product data as AI context.
Want early access? Join our MCP waitlist
Already have access? Learn how to connect
Read and Search Your Product Data
When a PM needs to reason about a product area - preparing for planning or answering a stakeholder question - they need to pull together everything relevant: features, epics, OKRs, feedback, related stories. In a large workspace, that means browsing, filtering, and opening items one by one, hoping nothing important gets missed. The context you bring to your AI tool is only as good as what you happened to find manually. With the Craft.io MCP connected, AI agents can search and read your workspace directly. Ask Claude to pull everything related to a topic and it searches across work items, portfolio items, and feedback, retrieves the relevant specs, and reasons about them with full context. No browsing, no switching tabs, no incomplete picture.
Create Items and Capture Feedback
Getting AI output into Craft.io has always required a PM in the middle: copy the text, open Craft.io, create the item, fill in the fields, repeat. The same friction applies to items’ updates - when needs to change items’ status, or edit descriptions. Likewise for feedback, signals arrive in calls, tickets, and Slack threads, and capturing them has always required someone to manually carry each one into the portal. With the Craft.io MCP, all loops close. A PM can instruct Claude to create structured work items directly from a brief, find all items related to a topic and update them at once, or route feedback from any external source into Craft.io automatically. A Gong call ends, the transcript is processed, and customer pain points appear in the feedback portal as structured entries - without anyone doing manual data entry.
User-Based API Keys with Scoped Access
Connecting an AI agent to a workspace is a significant step, and it requires appropriate controls. Craft.io's MCP implementation is built around user-based API keys - every connection is tied to a specific Craft.io user, and inherits exactly the permissions that user has in the platform. Beyond the baseline user permissions, each API key can be further scoped - restricted to specific workspaces, portfolios, or feedback portals, and limited to read-only or read-write access. Connecting the MCP server requires authenticating with a user-based API key.
Agent Actions are Tracked
The Craft.io activity log now shows who made the change - any change made through the MCP is logged under the connected user's name, with a "via Agent" label alongside it. Changes made through the API are marked "via API." Changes made in the product UI appear as they always have. The full history of what changed, who was responsible, and how it was done is visible in one place - giving teams the confidence to give agents write access without losing oversight.