Ask your AI for last week’s DAU and retention, straight from TA.
The metrics your team already defined in ThinkingData — DAU, retention, ARPU — over any window and grouped by up to three properties, plus the events and properties tracked in each project, answered inside ChatGPT, Claude, Codex or Cursor. Add your project key to toolfare once; every read is free.
1Connect your AI app
Pick the app you use. You do this once, and every API on toolfare is then available in it.
Add toolfare as a connector in Claude
Add to Claudehttps://api.toolfare.ai/mcp
- 1Open Customize → Connectors and choose "Add custom connector".
- 2Paste this URL as the remote MCP server URL.
- 3Leave the OAuth client id and secret under Advanced settings empty.
- 4Complete the browser authorization, then return to your conversation.
Custom connectors are available on Free, Pro, Max, Team and Enterprise. A Free account may hold one custom connector in total.
Claude’s own instructions2What to ask
Copy a prompt into your AI app once you’re connected. It picks the right one of the 5 tools and tells you the price first.
Your own metrics, over any window
Fetches up to ten of the metrics your team defined in ThinkingData over a recent or absolute window, grouped by up to three properties — the same numbers, with the same definitions, as in TA.
- “What was DAU by country for the last 14 days?”
- “Show day-1 retention by channel for last month.”
The metrics your team defined
The business metrics defined in a project, with the time granularities each one supports.
- “Which metrics are defined in our ThinkingData project?”
- “Is there an ARPU metric we can query weekly?”
The events a project tracks
Physical and virtual events tracked in a project, with their display names.
- “What events do we track in this game?”
- “Is there a purchase event, and what is it called?”
The properties behind an event
Event or user properties of a project, optionally for one event — what you can group a metric by.
- “What properties does the level_complete event have?”
- “Which user properties can we group retention by?”
Also available: The ThinkingData projects your toolfare account can read
1Create an API key
One key and one endpoint for every API on toolfare.
export TOOLFARE_API_KEY='tf_live_…'
2Every tool and its input
Send the tool name and its arguments as JSON; the reply is JSON too.
thinkingdata_query_metricFetches up to ten of the metrics your team defined in ThinkingData over a recent or absolute window, grouped by up to three properties — the same numbers, with the same definitions, as in TA.
thinkingdata_list_metricsThe business metrics defined in a project, with the time granularities each one supports.
thinkingdata_list_eventsPhysical and virtual events tracked in a project, with their display names.
thinkingdata_list_propsEvent or user properties of a project, optionally for one event — what you can group a metric by.
thinkingdata_list_projectsThe ThinkingData projects your toolfare account can read
Use it when
- You want a KPI your team already defined in ThinkingData, without opening TA.
- You want a metric broken down by country, channel or another property.
- You want to know which events and properties exist before asking a harder question.
Look elsewhere when
- You need ad-hoc SQL. SQL runs only after a person reviews it, and that step is not published through the MCP connection.
- You want installs and attribution by media source. That is your attribution account. Use this instead
ThinkingData directly, or through toolfare
Directly
- The ThinkingData Analytics interface
- Your own code against the Open API
- Exports to put game metrics next to ad spend
Through toolfare
- Ask in the chat you already use, in plain English
- The project key stays in toolfare, not in an agent
- Game metrics beside your ad accounts and attribution in one question
You get your defined metrics and schema, not ThinkingData Analytics. There is no SQL, no dashboard and no report saving through the MCP connection, and nothing is monitored for you between questions.
Limits
- The cluster is your own: every call spends its compute and shares a conservative per-connection quota.
- Times follow the project’s time zone, not UTC.
- Metric queries take no global filters in this version; narrow them with grouping instead.
- Older TA 3.x clusters may not expose metric listing.