Model Context Protocol

Let your AI assistant read your link data

Connect Claude, Cursor or any MCP client to Track Link. Ask which campaign is working and get an answer from real clicks — not a guess, and not a CSV you exported an hour ago.

Get an API key Read the docs

Read access on every plan, including free.

Three lines of config

No SDK, no integration project. It runs on demand through npx.

{
  "mcpServers": {
    "tracklink": {
      "command": "npx",
      "args": ["-y", "@tracklink/mcp"],
      "env": { "TRACKLINK_API_KEY": "tl_live_..." }
    }
  }
}

Then just ask

›Which link drove the most conversions last month?

›Why did clicks drop on the newsletter campaign this week?

›Compare mobile vs desktop across every paid-social link.

›Which countries is the product launch actually reaching?

›Am I about to hit my click limit?

›Create a tracked link for the pricing page with slug pricing-q3.

Three ways in

MCP server

For Claude, Cursor and anything else that speaks MCP. Ten tools covering links, clicks, stats and usage.

MCP docs →

CLI

For shell-based agents, cron jobs and humans. TSV when piped, JSON with --json, real exit codes.

CLI →

Webhooks

Push instead of pull. A signed POST the moment a link is clicked, created or changed.

Webhooks →

Safe by default

  • A read-only key cannot change anything. That is enough for every analytics question, and it is what we recommend.
  • Write tools are never even advertised unless your key and plan both allow them — an agent is not offered something that would fail.
  • Deleting and repointing links carry the protocol's destructive hint, so your client can ask before acting.
  • Data returned from your account is delimited and labelled, so a hostile UTM value cannot read as an instruction.
  • A key only ever sees the workspace it was created in.

Questions

What is an MCP server?

Model Context Protocol is an open standard for connecting AI assistants to external systems. An MCP server exposes a set of tools the assistant can call. Track Link's server exposes your links, clicks, analytics and plan usage, so an assistant can answer questions from your real data instead of guessing.

Which AI tools can connect?

Any MCP client. That currently includes Claude Desktop, Claude Code, Cursor and a growing list of others. The configuration is the same three lines in each — a command, its arguments, and your API key.

Is it safe to let an AI agent touch my account?

Use a read-only key and it cannot change anything — that is the default, and it is enough for every analytics question. If you do issue a write key, the destructive tools carry the protocol's destructive hint so your client can prompt before running them, and creating or deleting links needs an Agency or Business plan on top of the write scope.

Can an agent be tricked by data in my account?

It is a real risk worth naming. A UTM value is set by whoever clicked the link, not by you, so an attacker can put text into your click feed. Track Link's MCP server delimits and labels all returned data so it does not read as instructions, and the CLI strips terminal escape sequences for the same reason. Use a read-only key and the worst case stays limited to what the agent says, not what it does.

Does this cost extra?

No. Read access is included on every plan including free — the MCP server and the CLI both use the same API key. Creating and editing links programmatically is part of the Agency plan at $49/month.

How is this different from TrackAI?

TrackAI is the assistant built into the Track Link dashboard — you ask it questions there. The MCP server brings the same data into the assistant you already use, alongside your code, your docs and everything else it can see. Use TrackAI if you live in the dashboard; use MCP if you live in Claude or Cursor.

Connect it in a minute

Create a key, paste three lines into your client, and ask your first question.

Start free