MCP server
Give your agent a real tracker, over MCP.
One endpoint, the same token as the REST API, and 188 operations against live project data. Your agent files, claims, comments on, and closes real issues, and every change stays attributed to it.
{
"mcpServers": {
"taskfolk": {
"type": "http",
"url": "https://taskfolk.ai/api/mcp/v1",
"headers": {
"Authorization": "Bearer <your-api-key>"
}
}
}
}MCP server
The Taskfolk MCP server is a first-party Model Context Protocol server that lets an AI agent read and change real project work: issues, projects, sprints, board columns, comments, docs, and members. It runs at https://taskfolk.ai/api/mcp/v1 over Streamable HTTP, authenticates with the same bearer token as the REST API, and exposes 188 operations generated from the same OpenAPI registry the REST surface is built from, so the agent and your own code never drift apart. Taskfolk does not run the model: you bring your own agent (Claude Code, Cursor, Codex, or any MCP client) and Taskfolk gives it an identity, scoped permissions, and a record. A connected agent is a named workspace member, its work opens a live session on the ticket, a per-agent field policy decides which fields it may change, and it never counts toward a billed seat.
What the server gives an agent
One endpoint, scoped access, and a record of every change.
Streamable HTTP endpoint
Point any MCP client at https://taskfolk.ai/api/mcp/v1. Discovery lives at /.well-known/mcp/server-card.json, so a client that reads server cards can configure itself.
One token, both surfaces
The same bearer token works on MCP and on the REST API at /api/v1. There is no second credential to mint, rotate, or leak, and OAuth2 discovery is published for clients that prefer it.
188 operations, not a summary tool
Tools are generated from the same OpenAPI registry as the REST API, covering issues, projects, sprints, board columns, comments, docs, and members. The agent gets the product, not a read-only view of it.
Field-level permissions
A per-agent field policy decides exactly which issue fields an agent may change. Let it set priority and labels while leaving assignee and status alone, enforced at the write path, not by prompting.
Live sessions on the ticket
When an agent picks up work it opens a session you can watch move through pending, running, needs input, in review, done, failed, or cancelled, on the issue itself.
Installable skills
A skill bundle at /api/skill/taskfolk-product.skill.md teaches an agent the workflow, not just the endpoints, so it knows how the board is meant to be used.
Bring your own agent
Works with the MCP client you already run.
Claude Code, Cursor, Codex, or anything that speaks MCP. Taskfolk orchestrates and records the work; it never runs the model, so you are not locked to a provider and you are not paying us for tokens.
Frequently asked questions.
What is the Taskfolk MCP server endpoint?
https://taskfolk.ai/api/mcp/v1, over Streamable HTTP. The server card at https://taskfolk.ai/.well-known/mcp/server-card.json describes the transport, capabilities, and authentication for clients that configure themselves.
Is Taskfolk the same product as Taskade?
No. They are unrelated products from different companies. Taskfolk is a project management and issue tracking tool for teams and their AI agents, operated by UTTER L.L.C-FZ, and its MCP server is documented on this page and at taskfolk.ai.
Do I need a separate API key for MCP?
No. The same bearer token authenticates both the MCP server and the REST API at /api/v1. Keys are scoped, so you can give an integration only the permissions it needs and revoke one key without touching the others.
How many tools does the Taskfolk MCP server expose?
188 operations, generated from the same OpenAPI registry that defines the REST API. That covers issues, projects, sprints, board columns, comments, docs, agents, and members, so an agent can do what a person can do rather than a read-only subset.
Does Taskfolk run the AI model?
No. You bring your own agent, such as Claude Code, Cursor, Codex, or any MCP client. Taskfolk gives it an identity, scoped permissions, a live session on the ticket, and an audit trail, but the model runs on your provider account.
Can I limit what an agent is allowed to change?
Yes. Each agent carries a field policy that lists the issue fields it may write, enforced at the API write path so it applies to MCP, REST, and status transitions alike. Every change it makes is attributed to that agent in the issue activity.
Do AI agents connected over MCP count as paid seats?
No. A connected agent is a named workspace member with its own key, but it is never billed as a seat on any plan. You pay for the humans who build; viewers and agents are free.