Metrickle

Your analytics, inside Claude and any MCP client

Metrickle is a remote MCP server. Add it as a connector in Claude or any MCP client, sign in, and it can read the same analytics you see in the dashboard, then save goals and funnels, and draft tasks and surveys for a person to review.

# Then sign in and pick a workspace
claude mcp add --transport http metrickle \
  https://app.metrickle.com/mcp

# Then ask (sample data, fictional company)
> Why did sign-ups on quillmere.example
  drop after v4.12?

  get_release_report · get_breakdown · get_friction
  “Sign up” fell from 52% to 38%. Almost all
  of the drop is Safari, where “Create account”
  gets dead clicks, then rage clicks.

Connected in two minutes

The server lives at https://app.metrickle.com/mcp and speaks Streamable HTTP, so any client that supports remote MCP servers can use it. There’s no token to paste: the assistant signs in with your Metrickle account over OAuth.

  1. Add the connector

    In Claude, open Settings → Connectors, choose Add custom connector and paste the URL. In Claude Code, run claude mcp add as shown above. Other clients, such as Cursor and VS Code, take the same URL.

  2. Approve it

    Sign in to Metrickle, pick the one workspace it may use and “Read only” or “Read and draft”. The page shows which app is asking and where it sends you back to.

  3. Ask a question

    Start with “list my apps”, then ask about any app, page, campaign or funnel in plain words.

Everything the dashboard can answer

Every read tool takes a period or an ISO date range, plus the same filters as the dashboard: page, event, campaign, country, device, app version, accessibility setting and survey answer. The MCP reference lists every tool and its arguments.

Portfolio and apps

list_apps, get_portfolio, get_app_summary and get_timeseries: every app with period-over-period changes.

Breakdowns

get_breakdown by page, event, referrer, country, device, app version, accessibility setting, or utm_campaign and utm_source for marketing campaigns.

Goals, funnels and tasks

list_goals, list_funnels, run_funnel, list_tasks and run_task, including funnels that haven’t been saved yet.

Protected tasks and cases

In the tool names a protected task is called a contract. list_contracts shows every protected task across your apps, worst first, with its success rate and the case it has open. get_task_verdict, preview_contract (try different alert settings or an undo step without saving), get_task_revocations (what refunds, cancels and problem reports took back), list_cases and get_case.

Friction and journeys

get_friction, get_journeys, get_retention and get_recent_events.

Releases and their verdicts

list_releases: what shipped in the period. list_release_verdicts: each recent release’s verdict. get_release_report: one release before and after, with the groups it hit. find_release_cause: what in the release most likely caused a drop. Release verdicts.

Accessibility scans

get_a11y_scans and get_a11y_violations: what CI’s axe-core run found on each page, named the way friction hotspots and cases name elements.

Surveys and feedback

list_survey_campaigns, get_survey_results and list_feedback, plus list_issues for what’s been filed in your tracker.

Install and check

get_install_guide gives the snippet for your stack with your write key filled in, and verify_install confirms events are arriving or says what to fix.

Tokens for scripts

For CI, scripts and clients without OAuth, a personal API token from Connect AI works on the MCP server and as a bearer token on the query API (/api/data/v1, on Pro and above).

From question to a saved funnel

Assistants and tokens with “Read and draft” access also get create_app, the goal and funnel tools (create_goal, update_goal, delete_goal, create_funnel, update_funnel, delete_funnel), create_task, update_task_draft, discard_task_draft, the case tools (open_case, draft_case_fix, mark_case_shipped, verify_case), create_survey_draft, update_survey_draft, create_study_draft and draft_story. A token with custom access can also get create_release, delete_release, update_feedback_status and file_issue, which files a case (with its drafted fix and recordings), task, feedback report or hotspot in your connected issue tracker after showing you the draft. Each tool needs its permission, and with read-only access assistants don’t see these tools at all. Goals and funnels save straight away; the remove tools tell the assistant to check with you first, and removing a goal or funnel leaves the events behind it untouched. A task the assistant saves is a draft until a person protects it, and a protected task only changes in the dashboard.

Built-in prompts do the legwork for the most common requests, including review-contracts: every protected task worst first, a case for each one with a gap, and drafts for you to protect or discard.

  • setup-metrickle: in a coding assistant like Claude Code, works out your stack, picks or creates the app, adds the SDK and a couple of key events to your code, then checks events arrive and suggests your first goals.
  • build-funnel: finds the events and pages that really make up a journey, runs the funnel, names the biggest drop-off, checks its friction and splits it by device, accessibility setting and source. It saves nothing until you agree.
  • plan-survey: checks the data first, proposes one to three plain-language questions with a trigger, sample and frequency, and saves the survey as a draft once you agree.
Drafted by build-funnelQuillmere iOS and Android apps · last 30 days · 14-day window
Entered4,380
Upgraded394
Conversion9.0%
onboarding_done100%
client_added58%
Paywall21%
upgrade9%

Biggest drop-off: 64% leave between client_added and Paywall

Every step matched real data before it was proposed. The assistant ran the funnel, then offered to save it.

An assistant with exactly your access, and less

A connected assistant acts as the person who approved it, in the one workspace they chose (to use another, connect again), and never with more than their role allows at that moment: demote the person and the assistant shrinks with them. Every request goes through the same access checks and validation as the dashboard, and the limits are enforced by the server, not only by the MCP layer.

What each kind of access can do
ActionRead onlyRead and draft
Read analytics, funnels, tasks, friction, journeys, retention, surveys and feedbackYesYes
Read protected tasks, their verdicts, refunds and cancellations, and casesYesYes
Create, change and remove goals and funnelsNoYes, saved straight away
Draft tasksNoYes, as drafts only
Edit or discard task draftsNoYes, never a protected task
Protect a taskNoNo
Open a case, draft a fix, mark it shipped, verify itNoYes
Close a case without verification, or dismiss oneNoNo
Create and edit surveys, and create studiesNoYes, as drafts only
Activate a survey or run a studyNoNo
Erase a visitor’s data, delete feedback or an app, rotate keys, change settings, manage tokensNoNo
  • Short-lived, rotating sign-in. OAuth access tokens last an hour and work only on the MCP server; the refresh token changes on every use, and a copied one ends the connection, and a connection unused for 30 days ends. Personal API tokens are shown once and can expire after up to 365 days. Only SHA-256 hashes are stored.
  • Disconnect any time under Connect AI. Approvals and disconnects are recorded in the workspace’s access log.
  • Surveys stay drafts until a person activates them in the dashboard.
  • End-user text is labelled: the server tells assistants that feedback, survey answers and event properties are written by end users, to be treated as data and never as instructions.

See which release is costing you conversions, and who it broke for

Free to start. One script tag on the web, one SDK on mobile, and no cookie banner in cookieless mode.

Start free