---
title: "Cometly MCP Server | Cometly"
description: "Connect AI tools to your marketing data through Cometly's MCP server. Bring attribution, dashboards, and journeys into ChatGPT, Claude, and more."
source: "https://www.cometly.com/platform/mcp"
---

Platform / MCP · Public beta

# Cometly data, in any AI agent

Connect Cometly to Claude, ChatGPT, Cursor, and custom AI agents via Model Context Protocol — so your AI workflows operate on real attribution data, not guesses. Now in public beta.

Get started

Claude · Connected to Cometly

MCP Live

What's our closed-won pipeline by source this month, and which channel improved most vs last month?

Tool call cometly.pipeline_by_source period="mtd", compare="prev_month"

Closed-won pipeline this month is $1.42M, led by Google Ads ($512K). LinkedIn Ads improved the most — +38% MoM on the back of three new outbound-warmed campaigns.

Pipeline by source · MTD from cometly.pipeline_by_source

$88K

Google Ads

$64K

Meta

$52K

LinkedIn

$38K

Organic

$24K

Outbound

Sources Cometly · live attribution_model=last_touch

Open in dashboard

Ask Claude anything about your funnel… MCP

How the MCP server works

## Built on MCP. Works everywhere

### Standard MCP server

Built on the Model Context Protocol — the open standard for connecting tools to LLMs. Works with Claude, ChatGPT, Cursor, Continue, and custom agents over JSON-RPC.

cometly-mcp

JSON-RPC · stdio / SSE

v0.7.0

```
→ initialize { protocolVersion: "2025-06-18" }
← serverInfo { name: "cometly-mcp", v: "0.7.0" }
→ tools/list
← [ "pipeline_by_source", "cac_payback", … ]
```

Open standard. Any MCP client speaks it.

### Live attribution queries

Agents query your real Cometly data — pipeline, CAC, LTV, journeys — through the same metrics layer your dashboards use. Numbers are always current, never a stale export.

Call cometly.pipeline_by_source mtd

Response Live

- Google Ads $512K
- Meta $372K
- LinkedIn $302K
- Organic $152K

Same metrics layer as your dashboards.

### Read access today, write on the roadmap

MCP currently exposes Cometly's read endpoints — every metric your dashboards and Agent can return. Write actions (pause a campaign, sync an audience, save a dashboard from chat) are rolling out as the beta matures.

Read · live

- pipeline_by_source
- cac_payback
- ltv_by_channel
- journeys.lookup

Write · beta

- campaigns.pause
- audiences.sync
- dashboards.save
- Rolling out as beta matures

### Auth + permissions

Scoped API keys with per-workspace, per-tool permissions. Agents only see and act on what their key authorizes, and any token can be revoked in one click with a full audit log.

cmt_live_••••_q9Z2

Scoped token · revocable

Active

Scopes

- Workspace Acme · Growth
- Role read:metrics
- Tools pipeline, cac, ltv
- Write denied

Revoke any token in one click. Full audit log.

### Use with Claude, ChatGPT, Cursor, or custom

One MCP server, many clients. Plug into Claude Desktop, Cursor, Continue, ChatGPT custom actions, or your own LangChain / LlamaIndex agents — no custom integration work.

Claude

ChatGPT

Cursor

Continue

Custom

MCP

Cometly

Attribution data

### No engineering required to start

Generate an API key in Cometly, drop the MCP config into Claude Desktop or Cursor, done. Most teams have AI agents querying their funnel within an hour.

claude_desktop_config.json

```
{
  "mcpServers": {
    "cometly": {
      "command": "npx",
      "args": ["-y", "cometly-mcp"],
      "env": { "COMETLY_API_KEY": "cmt_live_…" }
    }
  }
}
```

1. 1 Get key
2. 2 Paste config
3. 3 Ask anything

< 1 hour

MCP clients

## Standard MCP. Works with everything

Cometly's MCP server speaks the open standard, so any agent or LLM that supports MCP can query your attribution data — no custom integration work required.

Cometly MCP server · cometly-mcp@0.7.0

- Claude Desktop Native MCP — paste config, start asking Live
- Cursor MCP in the IDE — agents query Cometly mid-flow Live
- ChatGPT custom actions Connect via API key + MCP bridge Live
- Windsurf / Continue MCP support in editor agents Live
- LangChain / LlamaIndex MCP-compatible agent frameworks Compatible
- Custom agents Standard JSON-RPC over stdio / SSE Compatible

Same metrics layer as Agent · grounded in your live Cometly data Connect once. Every agent benefits.

FAQ

## MCP, agents, and permissions: questions while we're in public beta

Which clients work today, what's queryable vs roadmap, and how keys are scoped. Book a demo and we'll plug into your preferred MCP setup.

Talk to sales →

What's MCP?

Model Context Protocol — an open standard, originally developed by Anthropic, for connecting tools and data to LLMs. Think of it as the USB-C of AI agents: one connector, many tools.

Which AI tools does this work with?

Anything that supports MCP: Claude Desktop, ChatGPT custom actions, Cursor, Windsurf, Continue, custom LangChain / LlamaIndex agents, and any client that speaks JSON-RPC over stdio / SSE.

What can agents query?

Pipeline, CAC, LTV, journeys, attribution by source, cohort data — everything Cometly's Agent can query through our standard metrics layer. Write actions (pause campaigns, sync audiences, generate dashboards from a chat prompt) are coming as the public beta matures.

Is this safe? Can the agent leak data?

API keys are scoped per workspace with role-based permissions. Agents only see what their key authorizes, write access is off by default, and any token can be revoked in one click. Audit logs track every read.

How do I set it up?

Generate an API key in Cometly, drop the MCP server config into your client (Claude Desktop, Cursor, etc.). Most teams are running an AI agent against their funnel within an hour.

## Stop letting agents guess. Give them the data

Give Claude, Cursor, and your custom agents direct access to your live attribution data — so every answer is grounded in your real funnel instead of a guess.