> /mcp
MCP servers
Usage: /mcp [reconnect|enable|disable [<server>|all]]
purpleprint
Remote HTTP · OAuth required · alwaysLoad truefilesystem
Connected · local stdiogithub
Disabled for this workspaceGot an idea? This coach turns it into a full product design through conversation. Paste → log in → start — that's the flow.
/purpleprint:start (or natural language “start PurplePrint” → pp_start). When the welcome shows, reply “let's start” and Phase 0 begins.Already have writing·notes·Notion? Paste them in too — it pulls the core and backfills what's missing.
Here's how the first conversation flows:
It's a demo — the real thing runs on your idea, inside your tool. Stuck? Just say “again” or “simpler” and it fixes course.
Our server is remote HTTP MCP + OAuth. No API keys, no config secrets — pick your tool and paste the prompt to your agent, and it registers itself and walks you through login. One login lasts 30 days.
| Tool | Status | How |
|---|---|---|
| Claude Code | Supported | URL + alwaysLoad: true + /mcp Authenticate |
| Cursor | Supported | url in user settings + browser OAuth |
| VS Code | Supported | Add the Claude Code or Codex extension and it connects through that. Direct registration in user settings is in the tab below (Copilot's own MCP is being validated) |
| Codex | Supported | codex mcp add/login + natural-language pp_* |
| Antigravity (Google) | Supported | config serverUrl → Refresh → Authenticate → paste code → restart (details below) |
| Devin Desktop · Cascade (formerly Windsurf) · Devin · Lovable | Coming soon | url + OAuth confirmed in docs · connection being validated (Devin needs org permission) |
| Emergent | Exploring | url-based + DCR OAuth (validating) |
Pick your tool's tab and paste the prompt straight to your agent. It registers and walks you through login.
Add the PurplePrint MCP to Claude Code.
Server name: purpleprint, transport: http, URL: https://purpleprint-mcp.purpleprintai.workers.dev/mcp.
Register it globally (user scope), and add alwaysLoad: true to mcpServers.purpleprint in ~/.claude.json
(this prevents /purpleprint:* slash commands from going missing in recent Claude Code IDE). If editing the file directly, save as UTF-8 no BOM.
Walk me through Authenticate in /mcp, and when done confirm that /purpleprint:start shows up.
⚠️ **Even if the /purpleprint:* slash commands never show up, it still works. Once connected, just say "start PurplePrint" in plain language — it runs the pp_start tool.**
The slash list is a convenience, not the entry point. Don't reinstall or get stuck on it (Ctrl+Shift+P → "Developer: Reload Window" or an editor restart sometimes brings it back, but it's fine either way).
If auth is needed and no prompt appears, disable and re-enable purpleprint, then check again. If it already shows "connected", auth is usually done — just say "start PurplePrint" to confirm. Add the PurplePrint MCP to this editor.
Server name: purpleprint, URL: https://purpleprint-mcp.purpleprintai.workers.dev/mcp.
Register it as a remote/http MCP in this tool's settings — globally (user settings, e.g. ~/.cursor/mcp.json or VS Code user settings), not just one project.
If an OAuth/authorize flow is needed, walk me through the browser login.
⚠️ **Even if the command list never shows up, it still works. Once connected, just say "start PurplePrint" in plain language — it runs the pp_start tool.**
The list is a convenience, not the entry point — don't get stuck on it (reload window, restart, or a new thread sometimes brings it back, but it's fine either way).
If auth is needed and no prompt appears, disable and re-enable purpleprint. If it already shows "connected", auth is usually done. Add the PurplePrint MCP to Codex.
Server name: purpleprint, URL: https://purpleprint-mcp.purpleprintai.workers.dev/mcp.
Register it as a streamable HTTP MCP in ~/.codex/config.toml (shared by Codex CLI/IDE) and complete the OAuth login.
On Windows PowerShell, if codex.ps1 is blocked by the execution policy, run it as cmd /c codex mcp ...
When done, verify with codex mcp list. It's fine if /purpleprint:start doesn't appear in the / menu —
just confirm that saying "start PurplePrint" lets you use the pp_start tool. Add the PurplePrint MCP to Antigravity as a global MCP.
Server name: purpleprint, URL: https://purpleprint-mcp.purpleprintai.workers.dev/mcp.
Register it via serverUrl in ~/.gemini/config/mcp_config.json, and check ~/.gemini/antigravity/mcp_config.json if needed.
After registering, go to Settings > Customizations > Installed MCP Servers, click Refresh, and complete Authenticate.
When done, confirm that saying "start PurplePrint" invokes the MCP tool. Press the login button and this screen appears — one tap with Google or GitHub:
PurplePrint
You need to log in to continue.
We'll verify with your account just once.
Stuck? See Troubleshooting below.
Client UI
> /mcp
Usage: /mcp [reconnect|enable|disable [<server>|all]]
purpleprint
Remote HTTP · OAuth required · alwaysLoad truefilesystem
Connected · local stdiogithub
Disabled for this workspaceSettings / Tools & MCPs
Ppurpleprint
Remote HTTPS · https://purpleprint-mcp.purpleprintai.workers.dev/mcp{
"servers": {
"purpleprint": {
"url": "https://purpleprint-mcp.purpleprintai.workers.dev/mcp"
}
}
} The MCP Server Definition 'purpleprint' wants you to authenticate to PurplePrint.
# ~/.codex/config.toml
[mcp_servers.purpleprint]
url = "https://purpleprint-mcp.purpleprintai.workers.dev/mcp" $ codex mcp add purpleprint --url https://purpleprint-mcp.purpleprintai.workers.dev/mcp
$ codex mcp login purpleprint
Opening browser for OAuth...
purpleprint · remote · authenticated
› 34 tools enabled
claude mcp add --transport http -s user purpleprint https://purpleprint-mcp.purpleprintai.workers.dev/mcp"alwaysLoad": true to mcpServers.purpleprint in ~/.claude.json — prevents missing slash commands./mcp → Authenticate on purpleprint → browser login~/.cursor/mcp.json or VS Code user settings):{ "mcpServers": {
"purpleprint": { "url": "https://purpleprint-mcp.purpleprintai.workers.dev/mcp" }
} } cmd /c codex mcp add purpleprint --url https://purpleprint-mcp.purpleprintai.workers.dev/mcp → cmd /c codex mcp login purpleprint/purpleprint:*) doesn't appear — say “start PurplePrint” and it uses the pp_start tool.This is the heart of PurplePrint — a 10-step coaching flow that turns a daunting one-line idea into a build-ready design. So you never start alone and stuck, the coach leads one step at a time, one question at a time. Each step stacks on the previous one's conclusion, building the “why→who→what→how” purpose chain with no gaps in the middle.
Try the tabs. The coach closes the gap first, you answer in plain words, and if you don't like it you can push back·debate — each of these is a way to use the product. It's not pressing “sounds good” like a general AI — you lead.
You never start staring at a blank screen. At every step the coach proposes options with a top pick and reasons, first.
How those usage modes actually play out — I followed a whole session, from Phase 0 to 9, on the one line “a cover-letter service.” Scroll and each step plays, and when a step finishes the artifact appears right there.
Below is the /purpleprint:<command> form for slash clients (Claude Code·Cursor). Tool clients (Codex) use the same features as pp_<command> — e.g. /purpleprint:status = pp_status.
/purpleprint:start | Start or resume — the first command you use. New? A warm onboarding; returning? The server remembers your last Phase·artifacts·next action and picks up right there. |
/purpleprint:Step progression (0–9) | You don't pick steps from a list — start, and the coach reads your state and leads you through context(0)·persona(1)·problem(2)·product(3)·UX(4)·wireframe(5)·check(6)·GTM(7)·dev design(8)·handoff(9). To jump to a specific step, just say “let's start from Phase 4”. |
/purpleprint:status | Shows at a glance which Phase you've reached, what artifacts you've made, and your next action. |
/purpleprint:list | Lists the artifacts you've made so far (context doc·design doc, etc.). |
/purpleprint:help-next-todo | Stuck? It points to exactly one next action from where you are. |
/purpleprint:projects | Got several ideas? See your project list to switch or create a new one. |
/purpleprint:save | Saves a checkpoint here. Pick up exactly there next time with start. |
/purpleprint:how-to-use | First time? A quick how-to for PurplePrint. |
/purpleprint:manual | A more detailed install·auth·troubleshooting manual. Reach for it when server-health doesn't resolve it. |
/purpleprint:server-health | “Connected but why isn't it working?” — self-diagnoses server reachability·source version·auth status and tells you per-tool recovery. |
/purpleprint:oauth | Briefs your login status·tier·remaining trial period. |
/purpleprint:handoff | Bundles your finished design artifacts into a package to hand a developer or AI directly. |
/purpleprint:my-growth | Analyzes your chat logs so far to measure how your builder capability has grown. It reads behavior signals only, not chat content (content-free). |
The PurplePrint design coach draws your design out from what you answer. But things like “how do competitors actually do it?” or “does this persona really exist?” are information that isn't in your head. The research & validation engines fill that blank with real investigation, so your design stands on evidence rather than guesses.
What's fundamentally different from general AI — General AI (ChatGPT·Claude, etc.) makes you ask for everything — “research the competitors” — and skips it if you don't. PurplePrint proposes, inside the Phase 0~9 framework, at the right moment: “now would be a good time for this research”. The coach flags what you're missing before you even know it's missing — that's the fundamental difference.
/purpleprint:hyper-research | Deeply researches competing products·tech·market → so you know “how others already solved this” and find a differentiator you won't just copy. (mainly Phase 2) |
/purpleprint:hyper-research-marketing | Researches competitors' channels·copy·content·community → so you start with evidence for which channels·messages land, instead of burning ad budget blindly. (mainly right after Phase 6) |
/purpleprint:hyper-research-persona | Validates with data whether your assumed persona actually exists → cuts the risk of betting on an “imaginary user,” and finds adjacent markets you missed. (mainly Phase 1) |
/purpleprint:hyper-scenario | Simulates your design as the flow a real user goes through → catches missing scenarios·break points before you build, reducing rework. (mainly right after Phase 5) |
/purpleprint:hyper-research-gtm | Researches search demand·competitor channels·launch cases → so you start with a strategy for where and how to get your first users at launch. (mainly Phase 7) |
How it's called — take the suggestion, call the command, and it runs on your AI agent's web capability (₩0 extra). By default the research itself is handed to a separate helper agent (Codex) — when research material piles up in the coaching thread, the design decisions that follow get dragged along by it. If you have a Codex subscription, the coach walks you through install and login, then runs the helper for you and only reads back the result (no need to open the Codex app). If you don't have one, the coach just does the research itself. Or call it yourself anytime. The “mainly Phase N” on each line is the best-fit moment, not the only one.
/purpleprint:hyper-review high tier
① Service design — persona·value chain·revenue·UX·security·team capability across 39 axes, recursing until zero findings. Answers 'is it a real product' at expert depth on the server. (mainly Phase 6)
/purpleprint:hyper-review-dar high tier
② Dev design — architecture·contracts·data models reviewed across 8 rotating lenses (DAR). Answers 'will the structure hold' before you write code. Only design docs go to the server, never the code. (mainly Phase 8)
/purpleprint:hyper-review-code-audit local
③ Code — using the earlier Review·DAR reports as the baseline, checks whether the code matches the design. Code never goes to the server (your AI agent audits locally · zero cost).
The three tiers (service design → dev design → code) are one review ladder. Each tier can go deeper with Ultra — lifting the round cap and recursing until findings hit zero (not a smarter AI, just a more relentless dig).
Here's how a Hyper-Review series run progresses (dev-design DAR example):
› pp_hyper_review_dar (dev-design docs as input)
# Hyper-Review DAR (dev-design architecture review) started
- jobId: dar-7f3a… · mode: dar (dev-design docs)
Check progress/result with hyper-review-status.
› pp_hyper_review_status (a few minutes later)
# Hyper-Review running
- Progress: round 3/8 (rotating deep review) — 7 findings so far
- Elapsed: ~4 min — usually 5–10 min
› pp_hyper_review_status (on completion)
# Hyper-Review complete
- Findings: 17 → 5 core issues
## Root cause
identity → value chain → async experience gap → missing billing cap → weak observability
## Core issues
1. [high] Undefined state transition when the billing cap is exceeded
2. [high] Metering atomicity = observation grade (not billing grade)
3. [medium] Hardcoded model prices …
## Next actions · team handoff
The report is saved to your local (docs/review/) — with teammate·AI handoff messages included.
A demo — in practice it runs on your design docs, across the three tiers (service-design Review · dev-design DAR · Code Audit) the same way. Only Code Audit keeps the code off the server — your AI agent audits it locally.
The design coach runs content-free — our server doesn't see what you're building or what you wrote. Exactly what lives where:
Your idea·chat·design docs (context doc·design doc·UX doc, etc.) all live on your computer only. Research-engine results too — they run on your AI agent's web capability, so the content never passes through our server.
Only progress signals like “which Phase you reached” or “how many times you used a command.” What the idea is, or what you answered — that content never goes. It's the minimum needed for resume·usage management.
Builder capability is measured from behavior signals (like which decisions you made yourself) without chat content. It doesn't look at what you said.
The design work itself needs no coding. Answer in plain words even without design terms — at each step it proposes 2~3 options and a top pick first, so you never start staring at a blank screen. Confused? Just say “again” or “simpler” and it fixes course right away. That said, PurplePrint runs inside an AI coding tool (Cursor, Claude Code, etc.) — it helps if you're already using one.
Paste them and it pulls the core, normalizes it, and backfills the missing “why·who·what problem.” You don't rewrite from scratch — it fills the gaps on top of what you have.
They stack as `.md` files in your local project folder (context·design·UX docs, etc.). Next time you `start`, the server remembers your last point and resumes right there.
Claude Code, Cursor, Codex, and Antigravity are supported. In VS Code, add the Claude Code or Codex extension and it connects through that. Devin Desktop·Cascade (formerly Windsurf), Devin, and Lovable are coming soon; Emergent is still being validated. If setup feels heavy, you can also run a review on the web. Detailed connection is in [Connect your tool] above, per tool.
Research·Scenario·Marketing·Persona·GTM run on your own AI agent, so PurplePrint adds ₩0. If the research is handed to a Codex helper agent, it runs on your Codex subscription (and if you don't have one, the coach does the research itself). Only Hyper-Review (39 axes), which uses a server LLM, is high tier.
Sign up and a 2-week free trial starts automatically, no card. After that, beta price $13/mo (list $20) — an add-on next to your existing AI subscription.
It's the minimum for account identity and trial/subscription management. Log in once with Google or GitHub and it lasts 30 days, so you don't repeat it.
No. Your chat·artifacts stay on your machine only. The server gets just progress meta like “which Phase you reached” (content-free). More in [Privacy] above.
That's normal. Codex is a tool surface — just say “start PurplePrint” and it uses the `pp_start` tool. Stuck? Diagnose with `pp_server_health`.
/purpleprint:*) don't showConnected (✓ Connected) but commands don't appear? In order:
~/.claude.json and check "alwaysLoad": true exists inside mcpServers.purpleprint. Add the line if missing. (Recent IDE may not surface slash in a fresh window without it.)/mcp → Reconnect on purpleprint. The command list reloads.Disconnect is usually unnecessary — the server always keeps the list open, so reload/Reconnect is enough.
Commands show but execution is blocked = token expired or not attached (the server is fine). Just log in again in your tool:
/mcp → purpleprint → Authenticate (or Reconnect) → browser loginpurpleprint → Authenticate. If it looks connected, hit Logout once, then log back in.MCP Servers, purpleprint → Login with PurplePrintOne login lasts 30 days. No need to repeat it.
You logged in, but an already-open session hasn't picked up the new auth yet (especially Codex). Open a new thread or restart the editor, then run again.
/) at all in CodexThat's normal. Codex is a tool surface — instead of slash, say “start PurplePrint” and it uses pp_start. Stuck? Diagnose with pp_server_health.
Still stuck? — server-health (self-diagnosis: server reachability·source version·auth status + per-tool recovery) and manual (detailed install·auth·troubleshooting) tell you the recovery path directly.