What is an MCP task board in plain language?
It is a task board that an AI assistant can query and update through a standard protocol, instead of one you have to describe or screenshot into a chat.
MCP task board
An MCP task board for Claude Code and Codex
MCP is the protocol that lets an assistant call your tools. Connect runnrr over MCP and Claude Code can read your board, pull a task's full context, and update cards as it works.
Short answer
Here is the short version.
An MCP task board is a task board an AI assistant can read and write through the Model Context Protocol. Connect runnrr as an MCP server and Claude Code, Codex, or Claude chat can list your tasks, find one by title, pull its full markdown context, create new cards, and move them through states, without you pasting anything.
That turns the board into shared memory rather than a screenshot. The assistant asks the board what is in flight and what a task is about, instead of asking you to describe it again at the start of every session.
Step by step
Set up an MCP task board for Claude Code and Codex
Step 1
Understand what MCP actually does here
The Model Context Protocol is a standard way for an assistant to call external tools. Without it, your board is something you describe to the model. With it, the board is something the model can query directly, which is why the assistant stops needing a briefing.
Step 2
Copy your connect command from runnrr
Open the Claude Code tab in your runnrr profile and copy the command. It carries your own API key, so treat it like a credential and do not paste it anywhere public.
claude mcp add --transport http runnrr https://runnrr.io/api/mcp \ --header "Authorization: Bearer <your-runnrr-api-key>"Step 3
Run it once in your terminal
Paste the command into your shell. Claude Code registers runnrr as an MCP server and the board tools become available in every session from then on.
Step 4
Ask the assistant to work from the board
Now the useful prompts change shape. Instead of describing a task, point at it. The assistant pulls the card, its markdown context, and the attached meeting notes on its own.
What is on my runnrr board right now? Pull the full context for "Send Kate pilot usage numbers", then draft it. Move that card to Review and note what you changed.
What you end up with
- The assistant reads your board directly instead of being told what is on it.
- Task context, meeting notes, and linked files arrive with the task, not separately.
- Cards can be created and moved by the assistant, so the board stays current during a session.
- The same board serves Claude Code, Codex, and Claude chat without a separate setup each time.
FAQ
Questions people ask before trying it
Which board tools does runnrr expose over MCP?
Listing tasks, summarizing the board by column, finding a task by title, pulling a task's full markdown context, creating tasks, updating a task's state or content, plus saving, listing, and sharing notes and logging decisions.
Does connecting over MCP send my board to the cloud?
The MCP endpoint is hosted, so tasks you expose through it are served from runnrr. Local agent runs and your markdown files stay on your Mac either way, and you can use the board without connecting MCP at all.
Does this work with Codex and Claude chat too?
Yes. Any MCP-capable client can connect to the same board. Claude Code is the most common, but the tools are not specific to it.
Who this is not for
- Anyone who wants zero cloud surface. The MCP endpoint is hosted, unlike local agent runs and the markdown files on disk.
- Teams needing shared boards or permissions. MCP access here is scoped to your own personal board.
- People not using an MCP-capable assistant. The board works fine on its own without ever connecting one.
Start now
Stop re-explaining your work to your AI
runnrr turns the follow-ups into cards your AI can finish. Free. No sign-up.
Download for MacmacOS 13+ · Web version coming soon
Related runnrr pages