COMING SOON · FOR TEAMS BUILDING WITH CODING AGENTS

Mainquest

Your agents forget what your team decided. Mainquest doesn’t.

Every agent gets the decisions that apply before it builds. Pull requests that break one are caught on the line. Open questions go to the person who owns the answer — not to whoever happened to be prompting.

We’re letting teams in a few at a time.
Mainquest’s Today screen: “Three things need you” — which name to publish under, code that charges students €25 where Pricing says €29, and a yearly-billing question — with what waits on others and what changed since yesterday.

Agents love a side quest.

Lars decided €290. An agent wrote €299. The tests passed and the review looked clean. Without Mainquest, customers find it first.

  1. The owner decides once

    DECIDED · B-4PRICING
    Yearly price for Pro €290 a year · two months free Decided by Lars, Finance · signed with his passkey

    Lars decides from one card on his phone. It lands in docs/decisions.md, next to the code.

  2. An agent goes on a side quest

    Three weeks later, working from a fresh chat, the agent guesses. Nothing it was given said otherwise.

  3. Mainquest flags the line

    mainquest check◆ code disagrees
    Line 42 sets proAnnual to €299. Lars decided €290 (B-4). Use 29000, or ask Lars to change the decision.
    Use 29000Ask Lars

    Caught in the pull request, on the line that caused it, before anyone is charged the wrong price.

01

Agents read before they build.

Connect Claude Code, Cursor or Codex once. Before each task the agent gets the few decisions that apply — not a 2,000-line CLAUDE.md. When it hits something nobody has decided, it asks instead of guessing.

The check in a pull request’s CI run: D-72, code disagrees — the code charges students €25 where Pricing says €29, at pricing/plans.ts line 14 — so mainquest check fails with exit code 1; a decision nobody confirmed is only a warning.
02

Drift is caught in the pull request.

The check runs in CI on every pull request. It reads plain files and uses no model, so it’s fast and it never makes things up. It fails only when code breaks a decision, and names the line, the decision and its owner.

03

See what’s decided, drawn out.

Pricing as price cards. Architecture as a map. Every decision you already have in markdown, laid out so the whole team can see it — and see what the code disagrees with.

Pricing drawn as price cards: Pro €29 a month, checked against Stripe each night; Pro yearly €290, waiting on Lars; Team €19 a seat a month, decided; every price includes VAT; and an exception for the app stores.
Mainquest in the terminal: Today’s list on the left, and on the right the open decision “Which name do we publish under?” with two numbered options and keys to decide, hand it to someone or ask.
04

Or never leave the terminal.

Everything in the app works from the keyboard: read, decide, hand off, sign. mainquest opens it; every command also takes --json for your scripts.

Your CFO doesn’t write markdown.

They shouldn’t have to.

When a price needs a decision, Lars gets one card in plain words, on his phone. No repo, no GitHub account, nobody else nudged. Agents can propose; only the person who owns the answer decides — and every decision is signed.

Finance · Lars
Prices, plans, refunds
Design · Sofie
Colours, type, brand rules
Product · Mia
What we build, and for whom

Nothing to migrate.

  • Your decisions stay in your repo

    Plain markdown in docs/decisions.md. Keep editing it the way you do now — Mainquest reads it on every commit.

  • Works with the agents you use

    Claude Code, Cursor, Codex or any MCP client. One command connects each.

  • Your code stays on your machine

    Reading and checking run locally and in your CI. No model, nothing uploaded.

  • Open source at the core

    The decision format, CLI, MCP server and check. Run all of it yourself.

What teams ask first.

Isn’t this just a better CLAUDE.md?

CLAUDE.md tells agents how to work, and any agent can edit it. Mainquest records what was decided and by whom, and checks the code still agrees. Keep your CLAUDE.md; Mainquest reads it.

How is it different from Spec Kit?

A spec builds one feature and goes stale once it ships. Mainquest keeps the decisions that outlive features true — across repos, with owners who aren’t in the repo. Use both.

Does my code leave my machine?

No. Reading and checking run on your machine and in your CI, with no model. Owner cards share decisions, owners and signatures — never source code.

When can I use it?

We’re letting teams in a few at a time so each one gets set up properly. Leave your email and we’ll write when it’s your turn.

Set out on the main quest.

The first teams get in soon. One command will show you what your team has already decided.

We’re letting teams in a few at a time.