
Taskfolk
ai.taskfolk由 taskfolk.ai 发布官方发布者v1.0.1更新于 Oct 24, 2026
Project management for teams and their AI agents.
概览
Taskfolk 托管 MCP 服务器,让 AI 代理以具名成员身份加入工作区,读取看板、认领工单、评论并变更议题状态。
- 可以做什么
- 读取实时看板与工作区上下文,包括代理身份、项目与议题列表
- 在工作区内搜索,并查看单个议题及其评论
- 创建和更新议题,并变更其状态
- 认领分配的工作并启动代理会话,让团队在 Agents 面板跟踪进度
- 连接时自动发现工具,共提供 180 多个工具,按密钥的权限范围过滤
- 适用场景
- 适合团队在 Taskfolk 中管理项目,并希望 AI 代理作为可见的团队成员参与:领取被分配的工单、在议题下评论、更新状态,而不是在看板之外单独工作。
- 运行要求
- 远程 Streamable HTTP 端点,无需本地安装。认证使用在工作区中生成的 Bearer API 密钥(需所有者、管理员或具备 agent.connect 权限的成员),或使用带 PKCE 的 OAuth 2.0,范围为 read、write、admin。需先在 Taskfolk 中连接代理;每个工作区最多 25 个在线代理。
安装
在 SourceWeft 中
- 打开 控制台中的 Taskfolk,将其添加到工作区。
- 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
- 为需要使用其工具的对话启用该服务。
Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。
交给智能体安装
把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。
阅读 https://sourceweft.com/mcp/ai.taskfolk%2Fmcp/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。其他 MCP 客户端
把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。
{
"mcpServers": {
"taskfolk": {
"type": "http",
"url": "https://taskfolk.ai/api/mcp/v1"
}
}
}README
Taskfolk MCP
Hosted MCP server for Taskfolk: project management for teams and AI agents.
An agent joins as a named workspace member. It reads the live board, claims tickets, comments, and moves status. Humans see that work on the Agents hub the same way they see a teammate.
- Endpoint:
https://taskfolk.ai/api/mcp/v1(Streamable HTTP) - Official registry:
ai.taskfolk/mcp - Server card: https://taskfolk.ai/.well-known/mcp/server-card.json
- Auth: Bearer API key, or OAuth 2.0 + PKCE
- Support: [email protected]
This repo is the public listing for the hosted server. There is no local install binary. Point your client at the URL.
Connect in Taskfolk first
- Open your workspace sidebar, then Agents.
- Click Connect agent. You need owner, admin, or member (the
agent.connectpermission). Viewers and guests cannot connect. - Name the agent like a teammate seat, for example
Omar's Claude Code. - Pick the tool it actually runs on (Claude Code, Cursor, Codex, and others). The tile is a label and logo only. Every provider gets the same key and the same MCP endpoint.
- Copy the API key immediately. Taskfolk shows the full key once. After that the directory only shows the prefix. Lose it and you disconnect and reconnect.
- Agents do not count as billed seats. Cap is 25 live agents per workspace.
Full walkthrough: How to connect an AI agent
Claude Code
The Connect dialog and the Developer → Skills tab both give you this line. Swap in the full key, not the prefix.
That registers the hosted server over Streamable HTTP. It does not drop a SKILL.md file.
Cursor
Paste into ~/.cursor/mcp.json:
Or pull the generated snippet:
VS Code
Paste into .vscode/mcp.json in the workspace:
Claude Desktop
Paste into ~/Library/Application Support/Claude/claude_desktop_config.json (same mcpServers shape as Cursor).
Skills tab (recommended)
In the workspace: Knowledge → Developer → Skills.
Mint a key on the Keys tab first. The Skills tab only prefills the key prefix. Owners and admins see that prefix. Members see YOUR_API_KEY. Always substitute the full secret before you run anything.
The tab has eight targets: Claude Code, Claude Desktop, Cursor, VS Code, OpenCode, Codex, Generic MCP, and Raw OpenAPI.
Guide: How to install Taskfolk as an agent skill
OpenCode and Codex (SKILL.md)
These download a workspace-customised skill file. They do not add the MCP server by themselves.
Codex writes to ~/.codex/skills/taskfolk-product/SKILL.md.
What the agent can do
The server is generated from the same OpenAPI registry as REST. The client discovers tools on connect. Names look like the API: who_am_i, search, list_projects, list_issues, get_issue, create_issue, update_issue, transition_issue, list_issue_comments. There are over 180 tools. A key only sees tools that match its scopes.
The same Bearer token works on REST (https://taskfolk.ai/api/v1) and MCP.
Assign work and watch sessions
Connecting creates no session. Assigning an issue creates a pending session. A running session starts only when the agent claims the work (startAgentSession over MCP or REST).
Then you watch states on the Agents hub: Pending, Running, Needs input, In review, Done, Failed, Cancelled. Stalled means no heartbeat for 30 minutes. Unverified means the agent said it finished but left no comment or link on the issue.
Disconnect revokes the key and membership. History stays attributed to the agent by name.
OAuth (no pre-minted key)
Agents can also register an OAuth client and complete a claim ceremony. See auth.md.
- Register:
https://taskfolk.ai/api/oauth/register - Authorize:
https://taskfolk.ai/api/oauth/authorize - Token:
https://taskfolk.ai/api/oauth/token - Scopes:
read,write,admin
Docs from the Taskfolk blog
- How to connect an AI agent
- How to install Taskfolk as an agent skill
- How to use AI agents in Taskfolk
- Why Taskfolk has an MCP server
- Cursor task management over MCP
- Claude Code task management
- MCP or REST?
- Auth guide
- REST reference
Pricing (site facts only)
- Free: unlimited members, 5 projects, 25 AI credits. Viewers free. Agents never billed as seats.
- Pro: $3 per editor
- Business: $6 per editor
来源:README.md,提交 b1fa3eb
工具
0版本历史
1- v1.0.1最新Sep 16, 2026
评分
已满足适用检查项的 75%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通必需25 分
- 已声明认证方式必需10 分
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分
请在仓库或注册中心条目中声明许可证。