# ChatStack

> AI PRD generator and real-time software cost estimator. An interview-driven 4-agent workflow produces structured user stories, functional and non-functional requirements, technical specifications, and a story-point-based estimate, optimised for AI coding tools and the Model Context Protocol.

ChatStack is a human-facing chat workflow on `https://www.chatstack.app/`. Prefer the APIs and Markdown below over scraping HTML.

## Agent paths

| Path | Auth | When to use |
|------|------|-------------|
| **Portal MCP / CRM REST** | Personal access token from Portal → Settings → API access | After a PRD exists — read/write packs, Improve/Build |
| **Public interview MCP** | None (guest token from `start_session`) | Run the 4-agent interview from an MCP client without a portal account |
| **Interview API (REST)** | Portal PAT **or** guest `interview_token` | Same flow without the iframe — `/api/agent/interview/*` |
| **`POST /api/agent/submit`** | `AGENT_API_KEY` (Bearer) | Skip the interview when the agent already has full structured artifacts (server-to-server only; never put this key in an MCP client) |
| **Browser UI (Playwright / Computer Use)** | None | Unofficial UI automation of the human interview — see `/agent-browser.md` |

Interview API: `POST /api/agent/interview/session|turn|action`, `GET .../status`, `POST .../submit-prd`. OpenAPI: `/openapi.json`. Public / guest traffic is rate-limited per network; 429 responses include a human `message`.

## Docs

- [Home (Markdown)](https://www.chatstack.app/index.md): Short product summary and links.
- [Browser agent notes](https://www.chatstack.app/agent-browser.md): Viewport, same-origin hooks, gate labels for UI automation.
- [What is ChatStack](https://www.chatstack.app/what-is-chatstack.md): Product overview and 4-agent workflow.
- [AI coding tools integration guide](https://www.chatstack.app/cursor-ai-prd-integration-guide.md): Personal token, `@chatstack/mcp`, file export, and project rules.
- [Popular Apps — MVP cost estimates](https://www.chatstack.app/popular-apps.md): Catalog of common app MVPs with full public PRDs (stories, requirements, tech specs, line-item costs), Year 0–5 TCO calculator, and notes on the post-interview hosting TCO card.
- [How much does it cost to build an app](https://www.chatstack.app/how-much-does-it-cost-to-build-an-app.md): Cost estimation logic, story-point model, 52% AI efficiency discount.
- [AI development services](https://www.chatstack.app/ai-development-services.md): Four tiers — free PRD, Requirements as a Service (MCP), consulting (workshops + enterprise AI), and AI-accelerated fixed-cost full-service delivery (nine-phase playbook).
- [ChatStack vs generic AI chat](https://www.chatstack.app/chatstack-vs-generic-ai-chat.md): Differences from ChatGPT/Claude for requirements work.
- [Vibe coding requirements workflow](https://www.chatstack.app/vibe-coding-requirements-workflow.md): Context discipline for AI-assisted development.
- [FAQ](https://www.chatstack.app/faq.md): Common questions.
- [About](https://www.chatstack.app/about.md): App Developer Studio credentials, AI-accelerated how-we-work loop, team, awards.

## API

### Marketing site (skip-interview submit)

- [Agent API discovery](https://www.chatstack.app/api/agent): Capabilities, auth model, rate limits, submit links.
- [OpenAPI 3.1 spec](https://www.chatstack.app/openapi.json): Submit endpoint and PRD artifacts.
- [AI plugin manifest](https://www.chatstack.app/.well-known/ai-plugin.json): Plugin descriptor for AI plugin hosts.
- Submit: `POST https://www.chatstack.app/api/agent/submit` with `Authorization: Bearer <AGENT_API_KEY>`. Request a key via `https://www.chatstack.app/speak-to-a-human`.

### Portal (read/write REST + MCP)

- Discovery: `GET https://www.portal.chatstack.app/api/agent` (operation catalog; same PAT for all agent CRM routes)
- Auth: `Authorization: Bearer <personal-access-token>` (Portal → Settings → API access). Membership-scoped only — never elevate via CRM super-admin on personal PATs.
- REST: read + write under `/api/agent/...` (stories, plan, requirements, techspecs, design, version create/lock/unlock, Improve, Build). Original quote versions (typically v0) are view-only for Improve/Build/lock.
- Journey: interview MCP → `submit_prd` → portal account → mint PAT → portal MCP. Full recipe + JSON: `/cursor-ai-prd-integration-guide.md`.
- MCP **portal**: `@chatstack/mcp` **stdio** (`CHATSTACK_MCP_TOOLSET=crm`) or hosted `https://chatstack-mcp-uiidp6d6gq-nw.a.run.app/mcp` (+ staging twin `…-staging-…`) with `Authorization: Bearer cst_…`. Env: `CHATSTACK_API_URL=https://www.portal.chatstack.app`, `CHATSTACK_TOKEN`. Never put `AGENT_API_KEY` in MCP config.
- MCP **public interview**: hosted staging `https://chatstack-mcp-interview-staging-uiidp6d6gq-nw.a.run.app/mcp` (no PAT). Call `start_session` first; tools: `start_session`, `send_turn`, `send_action`, `get_interview_status`, `submit_prd`. Production interview host ships with prod MCP CI (`chatstack-mcp-interview`); until live, use staging interview against staging marketing.
- Portal tools: packs, artifact writes, Improve (`get_improve_outputs` / `run_improve_agent`), Build (`get_build_prompt` / `run_build_prompt`). **Improve/Build generate** may take several minutes: async 202 then poll — do not treat silence as failure. Check `agents[].status` (`available` vs `in_development`) before `run_improve_agent`.

## Machine-readable

- [Full reference](https://www.chatstack.app/llms-full.txt): Dense product and schema reference.
- [AI dataset descriptor](https://www.chatstack.app/ai-dataset.json): Dataset and schema metadata.
- [Sitemap](https://www.chatstack.app/sitemap.xml): All pages and machine-readable assets.
- [robots.txt](https://www.chatstack.app/robots.txt): Crawler policy.

## Optional

- [Speak to a human](https://www.chatstack.app/speak-to-a-human.md): API keys, enterprise, full-service development.
- [NDA and terms](https://www.chatstack.app/non-disclosure-and-tcs): Legal (HTML only).