Automation & Workflows · Blog Post
monday.com Agent Skills: the official playbook for agents working on your boards
An MCP connection lets an agent touch your monday.com account. It doesn't teach it how boards, column values, docs, and forms actually behave. These six skills, written by monday.com, fill that gap.
By Mehadi Hasan 8 min read
What it is
mondaycom/skills is monday.com's own repository of Agent Skills, the folder-plus-SKILL.md format Anthropic open-sourced. The README calls them official skills that "equip any skills-compatible agent to manage boards, items, docs, and forms correctly". Every SKILL.md lists author: monday.com and an MIT license.
The skills don't call the API themselves. They are instructions that sit on top of the monday.com Platform MCP, the hosted server at https://mcp.monday.com/mcp, which uses OAuth, so you don't paste API tokens anywhere. If you haven't connected that server yet, my walkthrough on connecting Claude to monday.com with MCP covers setup, limits, and security. The MCP gives the agent tools like create_item and get_board_info. The skills tell it the order to call them in and which mistakes to avoid.
The repo is young and small. It was created on May 19, 2026 and last pushed on November 10, 2026. As of November 17, 2026 it has 6 GitHub stars, so it's far less visible than the unofficial monday skills that turn up in skill directories (more on those below).
What's in the repo
There are six skills under skills/. Three of them keep long reference material in a references/ folder that the agent only opens when it needs it. Every skill is still pre-1.0:
task-management
Version 0.2.1
Create, update, assign, and comment on items. Its main job is getting column values right: status, date, people, email, phone, link, checkbox and more each need a specific JSON shape, with a full reference file behind it.
board-setup
Version 0.2.2
Designs and builds a board (groups, column types, starter items), preferring one create_board call with a free-text creationPrompt, then checking the result and fixing group order, column types, and labels.
project-status-report
Version 0.1.4
Reads one or more boards and reports overdue items, blockers, workload, and recent activity, using monday's MCP chart, battery, and table widgets where the client supports them.
docs-collaborator
Version 0.2.2
Creates, reads, edits, and comments on monday docs, including live column-value embeds and mentions. The largest skill, with seven reference files on block operations, create modes, and errors.
forms-builder
Version 0.1.2
Builds and edits WorkForms that turn submissions into items on a backing board, with a question-type mapping table and notes on plan limits.
onboard-to-monday
Version 0.1.0
First-run onboarding for a new account: proposes one tailored board, builds it, adds a single starter automation, and hands over with one thing to try.
A GitHub Actions workflow, skill-size.yml, fails any SKILL.md over 24,000 bytes or 500 lines, with a body of 3,000 words or more, or with a description over 1,024 characters. The comments explain why: Google Antigravity only loads the first 24,000 bytes, and Google's Gemini partner plugin scanner rejects files over the other limits. So the repo is written to work in several clients, not just Claude.
How it works
Each skill's description says when to use it and also when not to, naming the skill that should take over. board-setup tells the agent to use task-management for editing items on an existing board. project-status-report hands off to docs-collaborator when you want the report written up as a doc. That keeps the six skills from overlapping when a request could fit more than one.
The bodies read like notes from someone who has watched agents fail on real boards. A few examples:
- Read the board first. task-management says to call
get_board_infobefore any create or update, because without the real column IDs and status labels, "column value writes will fail silently or set incorrect values." - Don't trust search ranking.
get_board_items_pagewith a search term returns the closest matches even when nothing matches. The skill says to check the name before writing, so the agent doesn't update the wrong item. - Check the generated board. board-setup notes that
creationPrompt"gets most of it right but not all". The agent then checks group order (new items land in the top group), column types, labels, and starter items. It also watches for automations nobody asked for, such as monday AI rules that use AI credits. - Don't retry blindly. If
create_boardtimes out, the agent searches for the board before trying again, because sending the call twice creates duplicate boards. - Know the platform limits. forms-builder starts by stating that WorkForms only create new items and can't update existing ones. docs-collaborator says never to edit a doc block it hasn't read, because block IDs only come from
read_docs.
monday.com also publishes a second, separate skill on its own domain: general-monday-gql-api-usage (version 1.2.0). It's one long file of about 760 lines. It recommends the MCP server first and raw GraphQL with a MONDAY_API_TOKEN as a fallback. It includes example GraphQL operations, pagination, rate limits, and error handling, plus a long section of behaviour rules and acceptable-use guidance for agents. The same /.well-known/agent-skills/ index also lists a signup-for-agents skill. Use the GitHub repo for day-to-day board work through MCP, and the well-known file when your agent can't reach a remote MCP server and has to call the API directly.
Official vs. unofficial monday.com skills
Search a skill directory for "monday" and you'll find several options. Only the ones above come from monday.com:
- Official:
mondaycom/skillson GitHub, the SKILL.md files served frommonday.com/.well-known/agent-skills/, and the MCP server (hosted at mcp.monday.com, with a self-run version published to npm as@mondaydotcomorg/monday-api-mcpfrom themondaycom/mcprepo, MIT-licensed). - Composio's monday-automation: part of
ComposioHQ/awesome-claude-skills. It runs through Composio's own Rube MCP and Monday toolkit, not monday's server. Copies of it also show up in other people's skill collections. - Membrane's monday skill: part of
membranedev/application-skills. It runs through the Membrane CLI and needs a Membrane account.
The third-party skills aren't necessarily bad. Composio and Membrane are useful if your agent already uses those platforms for many apps. But every request then goes through another company's connector and credentials, and the instructions are written by people who don't build monday. If you only need monday.com, the official set skips that extra layer.
Installation
The README says the skills work with any agent that supports the Agent Skills standard, and names Claude Code, Cursor, OpenAI Codex, and OpenCode. The quickest route is the npx skills CLI:
npx skills
# Install all skills npx skills add mondaycom/skills # Install a specific skill npx skills add mondaycom/skills --skill task-management
Or clone the repo and copy the folders you want into your agent's skill directory: ~/.claude/skills/ for Claude Code, ~/.cursor/skills/ for Cursor, ~/.config/opencode/skills/ for OpenCode, and ~/.codex/skills/ for Codex.
Clone / copy
git clone https://github.com/mondaycom/skills.git cp -r skills/task-management ~/.claude/skills/
Then connect the MCP server. These skills don't do anything without it. monday's own SKILL.md gives this config and notes that the exact syntax for a remote server differs between clients:
monday.com hosted MCP
{
"mcpServers": {
"monday": {
"type": "http",
"url": "https://mcp.monday.com/mcp"
}
}
}
You sign in through OAuth the first time the client connects. The agent then works with your monday.com permissions, so connect the account you actually want it to change.
Using it day to day
You don't call these skills by name. You ask in plain language, and the descriptions list the kinds of phrases that should trigger each one: "what's overdue", "weekly update", "create a board for tracking", "build an intake form", "draft meeting notes", "mark as done". A typical session might look like this:
- "Give me a status report on the Q4 launch board." project-status-report reads the board structure, finds overdue and stuck items, checks recent updates for why they're stuck (or the Notes column if there are none), and includes the board URL in the summary.
- "Move the landing page task to Done and assign the partner email to Priya." task-management looks up the column IDs, sends
{"label": "Done"}rather than a guessed index, and confirms with the item link. - "Set up a request form for design asks." forms-builder creates the form, fixes the default "New Form" title, shows you the questions before you share it, and reminds you that the layout setting has to be changed in the monday UI.
The skills also tell the agent how to report back: plain language, item and board URLs, no raw JSON, and no list of which API tools it called. That matters if the person on the other end is an ops manager, not a developer.
Pricing
The skills are free and MIT-licensed. What they can do depends on your monday.com plan, and the skills say so. docs-collaborator notes that Free and Basic plans are limited to 3 docs per account. forms-builder says removing the "Powered by monday.com" branding from a form needs Pro or above. onboard-to-monday skips the starter automation if the plan or trial rejects it. You'll also pay for whatever agent you run them in, and any monday AI automations the board builder adds use AI credits, which is why board-setup tells the agent to point them out.
My take
If you connect an agent to monday.com, add these skills too. task-management itself calls column value formatting "the #1 cause of failures", and that and board structure are exactly what task-management and board-setup cover. The notes on things like search ranking, duplicate boards, and default group order are the kind you usually only learn after things break, and here they come from the platform team.
There are limits. Every skill is below version 1.0. The set covers boards, items, docs, forms, and reports, but there's no dedicated skill for dashboards, sprints, or CRM pipelines beyond what the general board skills handle. The skills only describe what the MCP can do: if a setting isn't exposed there, such as a form's layout, the skill tells you to change it in the UI. They also don't replace a real integration. Two-way syncs, webhooks, and scheduled jobs belong in something like n8n or a monday app, not in an agent conversation. If that's what you need, that's the work I do under monday.com automation.
Who it suits: monday.com admins and ops leads who already use Claude, Cursor, Codex, or OpenCode and want the agent to handle routine board work without breaking columns. Developers who need raw GraphQL should start with the well-known SKILL.md. If your agent already runs everything through Composio or Membrane, their monday skills will fit that setup better, but they aren't from monday.com.
Frequently asked questions
Is mondaycom/skills the official monday.com skill set?
Yes. The repo lives in the mondaycom GitHub organization, the README describes the skills as official, and every SKILL.md lists monday.com as author. monday.com also serves a separate general-monday-gql-api-usage SKILL.md from monday.com/.well-known/agent-skills/. Skills such as Composio's monday-automation or Membrane's monday skill are third-party and run through those companies' own connectors.
Do I need an API token to use these skills?
No. The skills run through the hosted monday.com Platform MCP at https://mcp.monday.com/mcp, which uses OAuth: you sign in when your client first connects. Only the separate well-known GraphQL skill uses a MONDAY_API_TOKEN, as a fallback for agents that cannot reach a remote MCP server.
How do I install them?
Run npx skills add mondaycom/skills to install all six, or add --skill task-management (or another skill name) for just one. You can also clone the repo and copy skill folders into ~/.claude/skills/, ~/.cursor/skills/, ~/.config/opencode/skills/, or ~/.codex/skills/. Then connect the monday.com MCP server, since the skills depend on it.
Which skills are included?
Six: task-management (items and column values), board-setup (building and checking boards), project-status-report (overdue items, blockers, workload), docs-collaborator (monday docs with live embeds and mentions), forms-builder (WorkForms intake), and onboard-to-monday (first-run setup for new accounts). All are MIT-licensed and still below version 1.0.