Skip to content

Tracwell MCP

The dashboard answers the question you knew to ask, and the question that matters arrives as a sentence in a conversation. How many trial signups came from the launch post, which step of the funnel loses the most people this week, whether the revenue attributed to paid search moved after the price change - each one is three clicks into a reporting tool, or one question when the analytics sits behind an MCP endpoint. Tracwell is a web and product analytics platform that exposes traffic, funnels, goals, revenue attribution and session timelines to an agent, with the connections approved per website by the workspace owner.

Server type:  remote Streamable HTTP, OAuth 2.1
Endpoint:     https://tracwell.app/mcp
Auth:         sign-in through the client, or Bearer YOUR_MCP_KEY for API-key clients
Registry:     app.tracwell/analytics 1.0.0, remote declared as the same URL
Scopes:       analytics:read, definitions:write, websites:write, offline_access
Access:       an active plan, including trials; MCP stops when the plan lapses
Probe:        keyless initialize, then both OAuth metadata documents

The evidence tier

A keyless initialize answered 401 with {"error":{"code":"UNAUTHORIZED","message":"Sign in through OAuth or provide an API key."}} and the header WWW-Authenticate: Bearer resource_metadata="https://tracwell.app/.well-known/oauth-protected-resource/mcp". That header is the difference between a wall and a door: a client that follows it can onboard itself.

The protected-resource document answered 200 and named one resource, one authorization server at https://tracwell.app/api/auth, and four scopes: analytics:read, definitions:write, websites:write and offline_access. The authorization-server document answered on both the RFC 8414 path and the issuer-suffixed path with the same content: authorization_code and refresh_token grants, code as the only response type, S256 as the only PKCE challenge method, a registration endpoint for dynamic client registration, introspection and revocation endpoints, none, client_secret_basic and client_secret_post as token auth methods, and the issuer response parameter supported.

That scope set is the security story and it is a good one. Reading analytics is one scope. Writing is split in two, and neither of the write scopes is about the data itself: definitions:write covers goals and funnel definitions, websites:write covers the site configuration. There is no scope that edits or deletes analytics events, which is what a reporting assistant should not be able to do.

Two honest limits on this entry: the tool list was not measured, because the endpoint answers 401 without a connection and no account was created for this pass. Everything about tool names and returns is the vendor's documentation, labelled as such.

What the vendor documents

From tracwell.app/docs/agents and the pricing page, read on 2026-11-12. A connection is added by URL to any OAuth-capable client, the user signs in, and only workspace owners can approve a connection. At approval you choose which websites the connection can see and what it can do; reading analytics is always included. Connections are revoked in Settings → MCP, which is the operator's kill switch when a client goes stale or a laptop is lost. An API key is accepted as an alternative for clients that cannot do OAuth, and the docs state that all MCP access requires an active plan, including trials and complimentary access - when a plan expires the dashboard history stays but MCP stops until renewal, a clean boundary between "your data is still there" and "the agent can reach it".

The Free plan allows one custom connector. The product is European-hosted analytics with campaign tracking, custom events, funnels, goals, revenue attribution and session timelines, and the docs publish DPAs and GDPR support.

Published pricing, USD, read the same day:

monthly events   monthly   annual   websites   seats   history
300,000          9         90       5          1       365 days
1,000,000        15        150      5          1       365 days
2,000,000        29        290      20         10      730 days
5,000,000        49        490      20         10      730 days
10,000,000       79        790      50         30      1460 days
20,000,000       149       1490     50         30      1460 days

API and MCP access is listed as included at every allowance, not as an add-on, and each workspace gets one 7-day trial with no card. That single line is the reason this entry is worth a page: analytics vendors commonly meter the API separately, and here the agent surface is part of the same plan ladder.

Why the scope design matters to an operator

Most "connect your analytics" arrangements are a read-only API key pasted into a tool, and the key sees every property the account owns. The arrangement here is the other shape: an owner approves a connection, names the websites it may see, and the grant is carried in a token with four scopes that cannot touch the event stream. An assistant asked why did conversion drop last week needs analytics:read and nothing else, and the connection can be built that way and revoked without rotating a credential the rest of the stack depends on.

Where it sits

Cookieless Analytics MCP answers the same family of questions from a privacy-first platform over its own OAuth flow; Analytics Legends MCP is market intelligence rather than your own traffic. This is the first entry here where the grant is scoped to named websites before any question is asked.

Does reading analytics require a write scope?

No. analytics:read is its own scope, and the two write scopes cover goal and funnel definitions and site configuration - not the events or the reports.

What happens when the plan lapses?

The documentation states the dashboard history stays and MCP access stops until renewal, so an agent connection is not a way to keep querying analytics after a subscription ends.

Can a connection be limited to one website?

Yes - at approval the owner chooses which websites the connection may see, and the scope list carries websites:write separately for configuration changes. Per-website approval is the vendor's stated behaviour; the token grant is the measured part.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing