Task UI MCP
A server-side web job runner with an MCP endpoint: Claude Code runs jobs and workflows, reads their results and answers notes
- 97 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent runs commands on the server through jobs and terminals, and workspaces hold secrets and SSH keys
- An MCP token carries the rights of the user the agent works as, so take the minimal role
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/task-uiDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs commands on the server through jobs and terminals, and workspaces hold secrets and SSH keys
- An MCP token carries the rights of the user the agent works as, so take the minimal role
This is third-party code. Review the repository files before installing.
What it does
Task UI looks at every folder under a projects root, detects the conventions it follows and shows the targets as jobs: atkins pipelines, Taskfiles, Makefile targets, scripts from `package.json` and `composer.json`, and a `scripts/` folder. A job runs as a session on the server, and you can attach to it from any window while it runs. Projects live in workspaces with their own home directory, secrets and SSH keys. Roles and permissions decide what each user sees, edits and runs. Claude Code works beside the jobs: a request can be put to it from a folder, from a run or from a workflow step, and the Summarize button runs a job and asks Claude to retell how it went. The MCP server at `/api/mcp` serves an agent the same jobs, runs, notes and prompts as a user with a token of its own. The tools include `list_jobs`, `run_job`, `run_workflow`, `create_workflow`, `get_run`, `list_notes`, `note_comment`, `ask_prompt` and artifact handling. A note mentioning `@name` is an instruction to that agent, and `task-ui channel` delivers such notes to an open Claude Code session.
Who it is for. Teams and solo developers who run builds and service scripts on their own server and want to hand launching and analysis to an agent.
Good fit when
- Jobs from a Taskfile, Makefile and `package.json` are wanted in one web interface with a run history
- An agent should start a build, wait for the result and read the end of the output
- Handing tasks to agents through notes mentioning `@name` and tracking their status is convenient
- Chains of jobs and Claude requests are assembled into workflows and run on a schedule
Not a fit when
- A hosted CI with built-in repository hosting integration is needed: Task UI is installed on your own server
- The server with terminals and secrets cannot be protected: access means running commands
- A one-off `task` or `make` run from a terminal is enough
Example request
Run the build job in the api folder, wait for it to finish and send the last 50 lines of output, and if it failed leave a note with the reasonMCP
- Transport
- http
- Authentication
- API key
Environment variables
TASK_UI_URL- Base link of the Task UI instance, used to build endpoint addresses and links in the interface.
Limitations
This is a server application installed from a Docker image or a binary, and the agent needs a separate user with MCP enabled and a token. A token from the interface lives 30 days and is shown once, and the CLI can issue a token with no expiry. The endpoint accepts only tokens of users with MCP access enabled, and the agent's rights are bounded by the user's role, allow lists and hours. Delivering notes into a live Claude Code session goes through a channel in research preview and needs a development flag. Modules are written in PHP and run inside the binary. The documentation is in English.
How to disable. Run `claude mcp remove task`, then turn off MCP for the user with `task-ui user mcp claude off` or the Disable button in the profile. A CLI token is removed with `task-ui token claude rm <name>`.
FAQ
How do I connect Task UI to Claude Code?
Turn on MCP for the user in the interface or with `task-ui user mcp claude on`, get a token and run `claude mcp add --transport http --scope user task <address>/api/mcp --header "Authorization: Bearer <token>"`.
What can an agent do through MCP?
The same as the user in a browser with their role: view jobs and runs, run jobs and workflows, read and write notes and artifacts. The viewer role is enough for reading, and running needs runner.
How do I issue an agent token with no expiry?
On the server with `task-ui token <user> <name>`: the token is printed once and only its hash is stored. A new token with the same name replaces the old one.
How do I run Task UI itself?
The README quick start is `task docker:run`, and the Compose file is described in the documentation section on running the server.
How do I deliver notes to an open Claude Code session?
Connect the channel as a stdio MCP server with `task-ui channel`, then start the session with the development flag `--dangerously-load-development-channels`. The feature is in research preview.
Discussion
No questions or reviews yet.
Related
MCP serversThe cronstable job scheduler exposes its jobs, run history and schedule debugging to an agent through MCP.
A Rust CLI and TUI: agents in containers, parallel tabs, TOML workflows from issue to PR and a scheduled agent squad
The Coder MCP server manages workspaces and templates, edits files, runs commands and drives Coder agent chats.
Claude Watch draws a localhost map of project logic, shows Claude Code tool calls live and answers questions
Diffinery plugins connect Claude Code and Codex to self-hosted Git, pull requests and CI on Cloudflare through an MCP server
A local agent runtime and MCP gateway with 298 tools: files, Git, processes, Chrome browser, Windows and Office
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Task UI (this card) | no | API key | 97 | |
| GitLab MCP Server (Yoda Digital) | no | API key | 65 | MCP for the GitLab API, whereas Task UI runs Taskfile, Makefile and package.json jobs on your own server |
| HomeLab Monitor | no | not required | 236 | MCP for watching a home server, whereas here are jobs, workflows and notes with roles and agent tokens |
README in short
The documentation presents the application as a web interface for running what the sources declare, with workspaces, project files, roles, schedules and notes. It describes Claude working beside jobs, the MCP server, PHP modules, the look and a table of documents: the usage guide by section, running the server, the API, security and development.