CheckumiDocs
DashboardNew list

Checkumi for agents (MCP)

Connect Checkumi to any agent that speaks the Model Context Protocol. Your agent can create and edit checklists, read them, open a check run, report each item by its number (A1, A2…), and hand the run to a person for review.

SERVER URLhttps://checkumi.com/mcpStreamable HTTP

Connect your agent

Pick your client and add the server URL. There are no API keys to copy: the first time your agent needs your account, your client opens a browser window and you sign in with Google.

claude mcp add --transport http checkumi https://checkumi.com/mcp

# then, inside Claude Code:
/mcp   →  checkumi  →  Authenticate
A browser window opens for Google sign-in. Skip it to use public lists only.

Signing in (OAuth)

Checkumi uses the standard MCP OAuth flow. Your client handles it; you only click through once.

  1. 1Agent asks for accessWhen a tool needs your account, the server replies “sign-in required” and your client opens the browser.
  2. 2You sign in and approveSign in with Google, check what the agent can do, and click Allow. See the approval screen →
  3. 3The agent is connectedIt shows up in Dashboard › Connected agents, where you can disconnect it anytime.
Public lists · no sign-inAgents can read public lists, and start and report runs on them, without signing in.
Private lists · signed inAfter you approve, the agent can read and edit your lists, including private ones, and save runs to your account, within the permissions you allowed.
Creating and editing need sign-in: a connected agent can change only lists you own or collaborate on, and only if you allowed “Create and edit your lists”. Without signing in, agents can still read public lists and run them.

Tools

search_checklistsquery: string, limit?: number
Find public lists, plus your private ones once you’re signed in.
get_checklistlist: slug | url
Returns groups and items with stable IDs (A1, A2, B1…) and each item’s note.
start_runlist, label?, mode?: "agent" | "agent_then_review"
Opens a check run and returns its ID and url. Signed-in agents report with their account; without sign-in it also returns a token and review_url.
report_itemrun, id: "A1", verdict: "pass" | "fail" | "unsure", evidence?: string
Records one result. Call it once per item; calling again overwrites.
finish_runrun
Closes the run. In review mode the status becomes needs_review and the owner is notified to review it at url.
get_runrun
Returns the run’s status, results so far, and any human overrides.
share_runrun
Returns review_url — a link that lets a person without an account check or review the run. Anyone with it can edit until the run is done, so call it only when handing off.
create_listtitle, description?, tag?, visibility?, groups: [{ title, items: [{ text, note? }] }]
Creates a list you own. It can be public or private.
update_listlist, title?, description?, tag?, visibility?, groups?
Changes the list’s details. Passing groups replaces the whole structure.
fork_listlist, title?
Copies a public list as a new public list you own, linked to its source.
delete_listlist
Deletes the list and its runs.
add_group · update_group · delete_grouplist, group: "B", title?, position?
Adds, renames, moves or removes a group. Groups are addressed by letter.
add_item · update_item · delete_itemlist, id: "A3" | group: "A", text?, note?, position?
Adds, edits, moves or removes an item. IDs are positional: re-read the list after moves and deletes.

A typical run

  1. 1Read the listget_checklist({ list: "seo-launch" })
  2. 2Open a runstart_run({ list: "seo-launch", label: "/pricing", mode: "agent_then_review" })
  3. 3Check and report each itemreport_item({ run: 16, id: "A3", verdict: "fail", evidence: "canonical → ?ref=nav" })
  4. 4Finishfinish_run({ run: 16 }) → status: needs_review
  5. 5A person reviewsOwner gets a notification and confirms or overrides in the run page

Example prompt

“Use Checkumi to run the seo-launch checklist against https://example.com/pricing. Report every item with evidence, mark anything you can’t verify as unsure, then finish the run so I can review it.”

Limits and errors

  • Size: up to 26 groups (A–Z) per list and 100 items per group.
  • Free plan: 5 new lists and 100 check runs a month per account, counted from your sign-up date. Runs started without signing in don’t count, but each list takes at most 60 of them an hour.
  • auth_required: the tool needs your account. Your client should open the sign-in page; if it doesn’t, reconnect the server in its MCP settings.
  • not_found: the list is private or doesn’t exist. Sign in with an account that can see it.
  • run_closed: the run was already finished; start a new one.
  • unknown_item: the item ID isn’t in this run (IDs are fixed when the run starts). Re-read it with get_checklist.
  • forbidden: you or this connection can’t do that, for example editing without “Create and edit your lists”, or sharing a run on a list you can’t edit.
  • conflict: the list changed since you read it. Re-read it and try again.
  • too_large, too_many_groups: the change would go over the size limits above.
  • quota_exceeded: this month’s lists or runs are used up. The message says when the quota resets; nothing was created.
  • rate_limited: too many runs on this list without sign-in. Sign in, or try again later.