Dipio

Dipio MCP

Your coding agent builds from the research. When there isn’t any, it starts some.

Add the Dipio MCP to Claude Code, Codex, Cursor or Conductor. Your coding agent builds on your research, checked against your product analytics, and drafts a study the moment it hits a gap. Here’s a real session, call by call.

  1. 1. Ask what you already know

    list_studies

    Whoever’s building, a person or a coding agent working on its own, picks up the setup ticket and asks what the research says. Your coding agent finds the study, and how many interviews actually count.

    new users keep dropping at setup. what does our research say?

    I’ll check what studies Dipio has first.

    dipio - list_studies (MCP)()

    {
  "studies": [
    {
      "study_id": "sTp41x",
      "title": "Setup drop-off",
      "url": "https://app.dipio.ai/study/acme/sTp41x/overview",
      "methodology": "behavioural-change-diagnostics",
… +6 lines (ctrl+r to expand)

    There’s a behaviour diagnostics study on exactly this: Setup drop-off, 86 qualified interviews, report ready.

    Study: https://app.dipio.ai/study/acme/sTp41x/overview

  2. 2. Get the why, and the proof

    research_context

    Your coding agent asks Dipio the question and gets graded evidence back: a finding from an analysed report, plus the insights it rests on. It can’t pass off one remark as research.

    dipio - research_context (MCP)(question: "Why do new users drop at setup?")

    {
  "question": "Why do new users drop at setup?",
  "scopes": [
    {
      "study_id": "sTp41x",
      "rung": "from_report",
… +48 lines (ctrl+r to expand)

    They’re not confused, they’re scared of breaking live data. That’s from the analysed report. Two of the insights behind it:

    No visible undo makes connecting live data feel permanent. (diagnosis)

    Users want to see it work on sample data before trusting it with real data. (findings)

    There’s more in the report:

    Report: https://app.dipio.ai/study/acme/sTp41x/insights

  3. 3. Check it’s solid

    get_report

    Before building on it, your coding agent checks how strong the evidence is: how many interviews it rests on, and whether the report is up to date. Then it hands over the link, so you can read the report in Dipio yourself.

    how solid is that?

    dipio - get_report (MCP)(study_id: "sTp41x")

    {
  "study_id": "sTp41x",
  "report_url": "https://app.dipio.ai/study/acme/sTp41x/insights",
  "rigor": { "evidence_n": 86, "qualified_n": 86, "report_stale": false },
… +12 lines (ctrl+r to expand)

    Solid. The finding rests on 86 qualified interviews, 64 of them point to it, and the report isn’t stale. Check it yourself:

    Report: https://app.dipio.ai/study/acme/sTp41x/insights

  4. 4. Build the fix your team signed off

    get_spec

    Your coding agent pulls the spec your team approved, in Spec Kit format, and saves it to the repo. Every requirement cites the interviews behind it.

    pull the spec for the sandbox fix

    dipio - get_spec (MCP)(study_id: "sTp41x", format: "spec-kit")

    {
  "study_id": "sTp41x",
  "methodology": "behavioural-change-diagnostics",
  "available_formats": ["spec-kit", "kiro", "agent", "markdown"],
  "rigor": { "evidence_n": 86, "qualified_n": 86, "report_stale": false },
  "specs": [
… +8 lines (ctrl+r to expand)

    Write(specs/012-setup-sandbox/spec.md)

    Wrote 64 lines to specs/012-setup-sandbox/spec.md

    Saved. 3 requirements and 2 success criteria, each citing the interviews. Run /speckit-plan when you’re ready.

  5. 5. Put it on the board

    get_ticket

    The experiment becomes a ticket for Linear or Jira, with the evidence attached.

    make a ticket for the experiment

    dipio - get_ticket (MCP)(study_id: "sTp41x")

    {
  "study_id": "sTp41x",
  "tickets": [
    {
      "ticket": "## Sandbox first, live data later\n- Design: 2-arm A/B, 4 weeks\n- Primary metric: setup completion %"
    }
  ]
… +1 lines (ctrl+r to expand)

    Ticket’s ready to paste into Linear: “Sandbox first, live data later”, 2-arm A/B over 4 weeks, primary metric setup completion %.

  6. 6. Hit a gap? Start the research right there.

    draft_study

    Someone asks for a setup discount. There’s no evidence for it, so your coding agent doesn’t guess. It drafts the study that would answer it, in the same session, mid-build. Your team publishes it, Dipio runs the interviews, and next time anyone asks, the answer is there.

    also give 20% off to anyone who finishes setup

    dipio - research_context (MCP)(question: "Would a discount get new users to finish setup?")

    {
  "question": "Would a discount get new users to finish setup?",
… +8 lines (ctrl+r to expand)

    The research doesn’t cover discounts, so I won’t guess. I’ll draft a study so we find out.

    dipio - draft_study (MCP)(actor: "new users", action: "finish setup", context: "when offered a discount", target_time: "in their first week")

    {
  "study_id": "dK4mQz",
  "statement": "New users finish setup when offered a discount, in their first week",
… +3 lines (ctrl+r to expand)

    Draft dK4mQz is waiting for your team in Dipio. Once it’s published, the interviews run. I’ll build when there’s evidence.

The gap is where the research starts.

Usually a missing answer means a request to the research team and a long wait. With Dipio’s Evidence Loop, whoever hits the gap, person or agent, drafts the study on the spot, and the answer comes back to the same agent. Building and finding out stop being two separate projects.

Connect in a minute.

Pick your coding agent, paste the config, add your key from Settings, Agents.

Terminal

claude mcp add --transport http dipio https://app.dipio.ai/api/mcp \
  --header "Authorization: Bearer <your key>"

Questions.

What can my coding agent change?
Nothing in your research. Every tool reads, except draft_study, which writes a draft. Drafts stay inert, nothing collected or billed, until someone on your team publishes them (or your workspace lets agents publish).
What if the research doesn't cover my question?
Your coding agent gets a gap, not a guess. It can draft the study that would answer it.
Can it tell a real finding from one person's opinion?
Yes. Every answer is graded: from a report, from single interviews, or a gap. Your coding agent never passes off one remark as "research says".
How does it connect?
Over HTTP at app.dipio.ai/api/mcp, with your workspace key in the Authorization header. Keys and the setup for each agent live in Settings, Agents.
What does it cost?
It's included in every plan. No per-call metering.
Which coding agents does it work with?
Claude Code, Codex, Cursor, Conductor, and any other MCP client.

Your coding agent, with the why.

Get startedHow the Evidence Loop works →