/
MCP Server Reference

MCP Server Reference

MCP Server Reference

This is the technical reference for the Watching That MCP Server. For setup instructions specific to your AI platform, see the AI & LLM Platforms section:

What is the MCP Server?

The Watching That MCP server lets you interact with your platform data through AI assistants. Instead of building queries manually in the Analysis module, you can ask questions in plain English and get answers drawn from your Watching That data (recent detailed data plus historical rollups).

"Show me fill rate by distribution partner for the last 7 days"

Available Tools

The MCP server exposes 6 tools:

Tool

What It Does

Read / Write

Tool

What It Does

Read / Write

current_user

Get your authenticated user info (ID, name)

Read

list_accessible_dss

Discover all data sources your account can access

Read

list_available_metrics

Get all metrics for a data source (code, name, description where available, scope, applicable databases). For full metric definitions, see the Metrics & Dimensions Glossary.

Read

list_available_dimensions

Get all dimensions for a data source (field, name, description, applicable databases)

Read

allowed_long_term_dim_combinations

Get valid rollup dimension combinations — prevents query failures on historical data

Read

search

Execute data queries — the core workhorse

Read

Your account may have access to more than one data source (for example a FreeWheel log feed and a GAM feed). Each data source has its own metrics and dimensions, so list them per data source and name the data source (or data sources) you want when you query.

The search tool

Accepts a full query structure: data source selection (one or more), date ranges (ISO 8601 or relative like today-7d), metric columns (including formulas), groupBy configurations (time intervals, Top N, specific segments), and filters with 16+ comparison operators.

What's Not Covered

The following are only available through the web application: column-level metric filters, conditional formatting, visualisation configuration, saved explorations, Reports module, Monitoring module, Inspect (individual sessions and raw events), and user/org administration.

Connection URL

https://mcp.<instance>.watchingthat.us/

Your Customer Success representative will provide the complete URL. See the platform-specific guides above for how to use it.

Example Use Cases

  • Daily briefing: "Give me yesterday's key numbers: revenue, fill rate and errors versus the day before"

  • Revenue analysis: "Show last month's revenue by channel with eCPM"

  • Fill rate monitoring: "Show fill rate by distribution partner for the last 7 days"

  • Top advertisers: "Top 50 advertisers by revenue for programmatic, last month"

  • Troubleshooting: "Why is eCPM for one of my channels significantly below the portfolio average?"

  • Week-over-week: "Compare this week vs last week revenue by channel"

Tips for Best Results

  • Be specific about date ranges — "last 7 days", "last month", or "yesterday" all work

  • Name your metrics and dimensions — use the names from the Watching That platform. Ask the AI to list what's available if unsure

  • Start broad, then drill down — high-level question first, then follow-up

  • Name the data source — if your account has more than one, say which data source you mean. You can query across several at once

  • Use formulas — ask for calculated metrics like fill rate, eCPM, or completion rate

Limitations

  • Read-only analytics — cannot create or modify reports, monitors, or configurations

  • No UI features — conditional formatting, visualisations, and widget layouts are web-only

  • Query scope — scoped to the data sources your account can access

  • Large queries — very large or very frequent queries can be slow. Contact support if you run into problems

  • Data window — detailed data covers about the last 7 days. For anything older, ask your AI assistant to use the daily rollups, which support fewer dimension combinations

  • AI-written answers — your AI assistant writes the explanation; Watching That supplies the data. Check key numbers in the platform before acting on them

FAQ

Q: Do I need a separate licence for the MCP server?
A: Contact your Customer Success representative to discuss MCP server access.

Q: Can I use it with AI clients other than those listed?
A: Our setup guides cover Claude and ChatGPT. Other AI assistants that support remote MCP servers over streamable HTTP with OAuth sign-in should work, but we haven't tested them.

If you are connecting an automated agent rather than an AI assistant you use yourself: the Watching That sign-in may ask you to confirm you are human, so have a person complete the first sign-in in a web browser. AI clients that register themselves with Watching That can use refresh tokens. Request the offline_access scope when signing in, keep the refresh token, and renew access at the token endpoint listed in your MCP server's /.well-known/oauth-authorization-server document instead of signing in again.

Q: Is my data secure?
A: Yes. Each person signs in with their own Watching That login (through your company's single sign-on, if it's set up), and sees only what they can see in the web app. The AI tool you use processes the answers under your agreement with that provider.

Q: Can multiple team members use it simultaneously?
A: Yes. Each person connects and signs in separately. Don't share one login between people.

Q: Error about invalid dimension combinations?
A: Not all dimension combinations are available on historical (rollup) data. Ask the AI to check allowed combinations, then adjust.

Q: The AI says the search returned nothing?
A: Every metric and dimension in one question must come from the same underlying database. Ask the AI to check which databases each field applies to, then drop or swap the field that doesn't match.

Still Need Help?