Your agents don’t need to persist. Your purpose does.
How do I connect Amplitude to ChatGPT?
Amplitude over MCP hands ChatGPT the observational layer — dashboards, charts, experiments, flags, metrics — client-agnostic as ever.
Direct answer: connect Amplitude's MCP server to ChatGPT. In ChatGPT, add it from the connector settings (developer mode); the tool's own docs carry the current path. Authorised, ChatGPT reads your project's dashboards and charts, experiments and their arms, feature flags, and the metrics layer — the assembled views, retrievable by asking.
The metrics, conversationally
How the north-star panel moved, what the test's arms show today, which flags are live — answered from the project. The chart was always right; getting to it was the tax.
Small cohorts tell confident lies
Worked case: your team set a cohort-size floor — no product conclusions from cohorts under two hundred users, after a triumphant retention 'win' evaporated when the sixty-user cohort grew up. Amplitude will chart a cohort of six without complaint; refusing to conclude from it is statistical humility you codified.
Through Unl the cohort read arrives with the floor applied: this segment holds 140 users — below your line, observation only, no conclusions; the evaporated win is the reason. Excitement gets audited before it ships a roadmap change.
Amplitude to ChatGPT reads the metrics; through Unl the metrics answer to the statistical floors you set — small-cohort enthusiasm gets caught at read time.
Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. It works today: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.
Read further
Questions people ask
What is Amplitude's MCP read surface?
Charts, dashboards, experiments, feature flags and metrics — the same across every MCP client, ChatGPT included.
Why put statistical floors in Unl rather than in team training?
Training decays and hires rotate; an agreed floor with its evaporated-win story arrives inside every cohort read, permanently. The rule teaches itself.
Does this replace an analyst?
No — it enforces what your analyst already begged for. The floor is their advice, agreed, made present at the moment of temptation.
What if a cohort never reaches your floor?
Then you never get a verdict from it, and that is the correct outcome rather than a stalled one. The temptation is to lower the floor until the number qualifies. If you do, record it as a new floor rather than an exception, so the next reader knows which standard the answer was judged against.
What this is
Your agents don’t need to persist. Your purpose does.
Save the thoughts, decisions and targets worth keeping, each with its reasoning, carried into every AI session the moment they matter. Unprompted.
Your whole AI world. What you decided at the epicentre. It works inside Claude, Claude Code, ChatGPT and Cursor, and connects in a couple of steps.
MCP native·You decide·Model agnostic·Your data
The right information
Connect Unl to bring the right information into the moment.
Your sources, read against the criteria you set.
The full product, open. Free to start. Paid plans from $10 a month.