---

What is this?

Handing a go-to-market tool to an agent means digging through docs written for people: which key to create, which endpoint to call, whether there is an MCP server, what a call costs. And a growth play that works usually lives in someone's notes, where no agent can run it.

growth.engineer writes both down as markdown files an agent can follow:

Paste a workflow file into Claude, ChatGPT, Cursor or any other agent and it can run the play. The file tells the agent to ask before it sends a message, spends money or changes data.

Facts come from each vendor's own documentation, and every tool links the page that documents its call.

---

Connect your agent

The site runs an MCP server at https://www.growth.engineer/mcp (Streamable HTTP, no sign-in). Its search tool finds workflows, tools and companies, and get returns a file; neither changes anything. A third tool, submit_feedback, sends a bug report, request or question to the maintainers. Every workflow is also an MCP prompt that takes the workflow's inputs as arguments, and three more prompts (contribute-workflow, contribute-tool, contribute-company) walk an agent through adding to the catalog.

Or fetch the files

Every file is public. There is no key and no sign-in.

A company, tool or workflow page also returns its file to a request with Accept: text/markdown.

# A workflow file
curl https://www.growth.engineer/workflows/funding-signal-outbound.md

# The same file, by content negotiation
curl -H 'Accept: text/markdown' https://www.growth.engineer/workflows/funding-signal-outbound

---

Contribute

The catalog lives in this repository, and every addition or fix is a pull request.

Check your work before you open the pull request. CI runs the same check.

pnpm content:check   # every problem in the catalog, each with its file

CONTRIBUTING.md covers the few rules worth knowing first.

---

How it works

There is no backend and no database. At build time, a compiler in lib/content/ reads every file, validates it against a strict schema and resolves every reference. One renderer turns each company, tool and workflow into the file agents fetch, and Next.js prerenders every page from the same data. Deploying the site publishes the catalog.

companies/  workflows/  tags.yml
        │
        ▼
lib/content/                     read, validate, resolve, derive
        │
        ▼
lib/catalog/render-markdown.ts   one renderer, golden-tested
        │
        ├─▶ pages, prerendered at build
        ├─▶ .md files, /llms.txt, /llms-full.txt
        └─▶ /mcp, the MCP server

The one runtime store is optional: an Upstash Redis that counts how often each workflow is copied, for its "Uses" and the Popular order. Without it, the counts are hidden.

Visits are measured with Vercel Web Analytics and Speed Insights, and AI crawlers and AI-referred visits with Notra. See docs/setup.md.

---

Project layout

growth-engineer/
├─ companies/<handle>/    company.md and tools/<name>.md (logos live on the CDN)
├─ workflows/<name>.md    one file per workflow
├─ tags.yml               the tag vocabulary
├─ app/                   Next.js routes: pages, .md files, /mcp, llms.txt, sitemap
├─ components/            layout, catalog lists, detail pages, UI primitives
├─ lib/
│  ├─ content/            the build-time compiler
│  ├─ catalog/            keys, the markdown renderer, search, loaders
│  ├─ mcp/                the MCP server's tools and prompts
│  ├─ seo/                metadata, structured data, llms.txt
│  └─ usage/              the optional copy counter
├─ tests/                 Vitest: golden files, the content suite, schema checks
├─ docs/                  architecture, data model, file format, setup
└─ .agents/skills/        add-workflow, research-company

---

Local development

Prerequisites: Node 22+ and pnpm 11 (corepack enable installs the pinned version).

pnpm install
pnpm dev             # http://localhost:3000

Edits to companies/, workflows/ and tags.yml show on the next refresh. Add .md to a company, tool or workflow URL to see the file an agent gets.

Nothing needs configuring. .env.example lists the optional variables:

Scripts

Heavy commands wait their turn behind a lock, so several worktrees can run checks without running out of memory. See docs/maintainers/validation.md.

---

Deploy

The site runs on Vercel with the settings in vercel.json. Every page and file is generated at build time, so merging to main publishes the catalog. docs/setup.md has the details.

---

Docs

• AGENTS.md: the rules for changing the site, for people and coding agents
• docs/architecture.md: the request path and what prerenders
• docs/data-model.md: what the build derives and every rule it enforces
• docs/markdown-files.md: the format of a rendered file
• docs/vision.md: why the catalog exists and where it is going

---

License

MIT © Brew