One URL for every project
https://mcp.togohq.ai/mcp
Togo’s MCP server speaks Streamable HTTP and signs you in with OAuth in your browser. There is no API key to copy. One connection reaches every project you are a member of, across your organizations, so each tool call names the project it is about.
- Claude Code: Connect Claude Code — one command and a sign-in.
- Cursor, Codex, VS Code and others: Other agents.
What your agent can do with it
- Set up tracking. The
setupprompt, or theget_setup_instructionstool, gives the agent the step-by-step install for your project, with its real write keys (Install Togo with your AI agent). - Check an install.
verify_trackingandverify_tracking_by_event_idsread a test run’s events within seconds, with what is wrong with each and the fix (Check that events arrive). - Read and keep the event registry. List events and their properties, define what an event means, merge two names for one event.
- Answer questions. Trends by day, week or month, funnels, metrics, and one person’s profile and recent events.
The MCP tools reference lists all of them, with the exact text your agent reads.
Read only, or read and change
When you sign in, you choose what this agent may do:
| Access | Tools it sees | Choose it for |
|---|---|---|
| Read only | Every read tool | An agent that answers questions |
| Read and change | Every tool, including the ones that change data | The agent that sets up and maintains your tracking |
Each agent you connect gets its own sign-in, so the one wiring tracking into your code can have Read and change while the one you ask questions in has Read only. Writes are also limited by your role in the organization that owns the project. See What a connection can see and do.
Questions that work
Which of my Togo projects is this app? Use the write key in the code.
How many people signed up each week for the last 12 weeks?
What share of people who viewed pricing started a checkout within 7 days?
Which events are sent under two names that look like the same event?
Every answer says which project it is about, what it covered and how fresh the data is. Live data is a few hours behind: a question about the last hour gets “not processed yet”, never a false zero (When your data shows up).
When a question is too big
A question that would scan too much, or a date range past its limit (365 days by day, 104 weeks by week, 36 months by month), is refused with a reason and a narrower question to ask instead — never answered from a sample or silently shortened. The agent can then ask the narrower one.