trackrev-mcp exposes TrackRev as Model Context Protocol
tools: 61 of them, over stdio. Point Claude, Cursor, or anything else that speaks MCP at it and
it can create tracking links, answer revenue questions, approve partners and reconcile
commissions against your workspace.
Every tool is generated from the same command registry as the CLI, which is
generated from the same API. One surface, three ways in — the numbers
always match the dashboard.
Install
trackrev keys create --scope secret.
Requires Node 20 or newer.
The tools
Each is named after the command it runs:trackrev links create is trackrev_links_create,
trackrev payouts mark-paid is trackrev_payouts_mark_paid.
Flags become typed parameters —
--days N is a number, --retarget on|off is an enum, a
repeatable --channel is an array — and a parameter the CLI documents as required is required in
the schema. A parameter that isn’t real is rejected by name: a model that guesses partnerId
is told the parameter is partner, rather than getting an unfiltered answer it believes was
filtered.
Analytics, affiliate and money tools need a paid plan, and fail with plan_required on a free
workspace. Link and account tools work on every plan.
What it deliberately won’t do
Destructive tools need confirm: true
Destructive tools need confirm: true
The 11 tools that cannot be undone — deleting a link, banning a partner, voiding a commission,
marking a payout paid — take a
confirm boolean that must be true. It is the MCP equivalent of
the CLI’s --yes, and the server enforces it as well as declaring it in the schema. Treat it
as an intent signal, not an authorisation system: the real gate is your client asking you to
approve the call.--read-only removes them entirely
--read-only removes them entirely
Run the server with
--read-only and every tool that changes data disappears from the list.
29 tools remain and nothing reachable can write. Useful for an agent whose job is to answer
questions about the numbers.No login or logout
No login or logout
Those rewrite the local key file and prompt for a secret. Which key is in play is decided by
whoever launched the server, not by the conversation.
No writing files to your disk
No writing files to your disk
The CLI can write a QR or a CSV export to any path you name. The MCP tools return the bytes
instead — “write this file wherever I say” is not a capability a link tracker needs to hand an
agent.
No creating payout batches, and no emails
No creating payout batches, and no emails
Creating a batch stays in the dashboard, as it does in the CLI: it applies per-group payout
floors and platform fees, and a second implementation would eventually pay someone wrong.
partners approve and payouts mark-paid change the record without sending the mail the
dashboard sends, so a retry loop can’t mail anyone twice — it will change the ledger twice.Flags
No shell, no MCP client?
The same surface is plain HTTPS at/api/v1 with the same key, and
the same commands are in the CLI for anything that can run a shell command.
