Growth Engineer

What growth-engineer does

growth-engineer turns a catalog of go-to-market resources into content that agents can search, retrieve, and follow. Catalog entries cover three types of material: companies and the ways agents can reach them, individual tools such as an enrichment function, and workflows that chain multiple tools toward a defined outcome.

Each entry is stored as a Markdown file. Company records describe the provider, access method, and credential requirements. Tool records identify an MCP function, CLI command, or API endpoint and link to the relevant vendor documentation. Workflow records describe their objective, expected inputs, setup requirements, and ordered steps. Workflows also tell an agent to request confirmation before sending messages, spending money, or changing data.

How it works

The hosted MCP endpoint is available at https://www.growth.engineer/mcp over Streamable HTTP and requires no authentication. The search tool locates companies, tools, and workflows. The get tool returns a catalog file, and both are read-only with respect to the catalog and external systems. submit_feedback sends a bug report, request, or question to the maintainers.

Every workflow is exposed as an MCP prompt whose arguments correspond to that workflow's inputs. Three additional prompts—contribute-workflow, contribute-tool, and contribute-company—lead an agent through adding catalog content. The same material can also be fetched directly from public Markdown routes, including /workflows/<name>.md, /tools/<company>/<name>.md, and /companies/<handle>.md. Aggregate indexes are available through /llms.txt and /llms-full.txt.

The repository compiles and validates the Markdown catalog at build time. A content compiler resolves references and a shared renderer produces pages, Markdown responses, index files, and the MCP catalog. Next.js prerenders the site from that generated data; there is no required backend database for catalog delivery.

Setup and configuration

For the hosted service, add the Streamable HTTP URL to an MCP-compatible client. Claude can use a custom connector, Claude Code supports claude mcp add --transport http, Codex supports codex mcp add, and Cursor can reference the URL in ~/.cursor/mcp.json. No key or sign-in is needed.

To run the project locally, use Node.js 22 or newer and pnpm 11. Enable Corepack, install dependencies with pnpm install, then start the development server with pnpm dev. The local site runs at http://localhost:3000.

Configuration is optional. NEXT_PUBLIC_SITE_URL controls the origin used in generated metadata and llms.txt. Redis variables enable workflow-copy counts, while GITHUB_TOKEN supplies the header's GitHub star count and NOTRA_GEO_TOKEN enables AI-traffic reporting. These services are not needed for the catalog or MCP endpoint to operate.

Tools and capabilities

• Search companies, tools, and workflows through MCP.
• Retrieve individual catalog files through MCP or public Markdown URLs.
• Run workflow prompts with their defined inputs.
• Use contribution prompts for new workflows, tools, and companies.
• Submit maintenance feedback through MCP.
• Browse tag-filtered content and full catalog indexes.

The MCP catalog is intended for discovery and instructions. It does not itself execute the vendor APIs, CLIs, or MCP tools described in catalog entries.