# Taskfolk [Health: Active]

**Category:** 💬 Communication  
**Repository:** https://github.com/taskfolk/mcp  
**GitHub Stars:** 0  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/taskfolk-2

## Description
Project management for teams and their AI agents.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "taskfolk": {
    "url": "https://taskfolk.ai/api/mcp/v1"
  }
}
```

## Documentation

## What the Taskfolk MCP server does

The Taskfolk MCP server gives AI agents access to project-management work in a Taskfolk workspace. Each connected agent appears as a named workspace member, allowing people to see which agent performed an action. Agents can inspect live boards, claim assigned tickets, add comments, and move issues through their workflow.

The hosted endpoint is `https://taskfolk.ai/api/mcp/v1`. There is no local executable to install. The server is generated from the same OpenAPI registry as Taskfolk's REST API, so its tools follow the API's naming and resource structure.

## How it works

A workspace administrator or another member with the `agent.connect` permission creates an agent from the Agents area, names it, and copies an API key. The key is shown in full only once. The agent then connects to the shared MCP endpoint using an Authorization bearer header.

The Taskfolk MCP server exposes tools dynamically when a client connects. A key only receives tools allowed by its scopes. Example tool names include `who_am_i`, `search`, `list_projects`, `list_issues`, `get_issue`, `create_issue`, `update_issue`, `transition_issue`, and `list_issue_comments`; the service provides more than 180 tools.

Assigning an issue creates a pending session, but a running session begins only after the agent claims the work through `startAgentSession` over MCP or REST. The Agents hub reports states such as Pending, Running, Needs input, In review, Done, Failed, and Cancelled. A missing heartbeat for 30 minutes marks a session stalled.

## Setup and configuration

Configure the hosted URL in a client that supports MCP over HTTP and provide the full API key in the Authorization header. The README includes setup formats for Claude Code, Cursor, VS Code, and Claude Desktop. Workspace-specific skill files are also available for OpenCode and Codex, but downloading a skill file does not itself register the MCP server.

OAuth 2.0 with PKCE is available when a pre-created API key is not suitable. The OAuth endpoints support `read`, `write`, and `admin` scopes. The same bearer token can also authenticate against the Taskfolk REST API at `https://taskfolk.ai/api/v1`.

## Tools and capabilities

The Taskfolk MCP server supports operations across Taskfolk's project-management API, subject to the connected key's scopes. Supported work includes:

- Identifying the connected workspace member
- Searching workspace data
- Listing projects and issues
- Reading individual issues and their comments
- Creating and updating issues
- Transitioning issue status
- Claiming assigned work and starting agent sessions
- Reporting activity that humans can review in the Agents hub

Agents are not counted as billed seats, and a workspace can have up to 25 live agents.

## Limitations and notes

This is a hosted service, not a repository that provides a local server binary. Connecting an agent does not start a work session; assignment and claiming are separate steps. Disconnecting revokes the key and membership, although prior history remains attributed to the agent's name.

The workspace must have an available agent connection, and viewers and guests cannot connect agents. Taskfolk shows a newly created API key only once; losing it requires disconnecting and reconnecting the agent. A session can be marked unverified when an agent reports completion without leaving a comment or issue link.

_Full upstream README: https://allmcps.com/mcp/taskfolk-2/readme_