Lathe

npm for MCP servers. Paste any URL, get a live MCP server, install it in one line.

Lathe turns any website into a tool your AI agent can use — in under 60 seconds. Powered by the Anakin scraping API.

What it does

You paste a URL (or describe a topic)
Lathe checks Anakin's Wire catalog — if the site is already supported, a server is created instantly
If not, Anakin crawls the site, infers the data schema, and generates a live MCP endpoint
You get a one-line JSON snippet to paste into Claude Desktop
Claude can now read that website as structured data

No code. No hosting. No maintenance.

Demo

Input:  https://news.ycombinator.com/jobs

Output: {
  "mcpServers": {
    "news-ycombinator-com-jobs": {
      "url": "https://lathe.dev/api/mcp/news-ycombinator-com-jobs"
    }
  }
}

Paste that into your claude_desktop_config.json, restart Claude, and ask:

"Any Rust jobs on HN this week?"

Claude calls the MCP server, Anakin crawls HN live, returns structured results.

Tech Stack

Layer	Tool
Framework	Next.js 16 (App Router)
Styling	Tailwind CSS
Database	Turso (libSQL — SQLite-compatible, hosted)
Scraping	Anakin API (Wire, Crawl, Agentic Search)
Schema inference	OpenAI `gpt-4o-mini` (fallback only)
Deploy	Vercel

Getting Started

1. Clone and install

git clone https://github.com/your-username/lathe.git
cd lathe
npm install

2. Set up Turso

Create a free database at turso.tech:

npm install -g @turso/cli
turso auth login
turso db create lathe
turso db show lathe          # copy the libsql:// URL
turso db tokens create lathe # copy the token

3. Configure environment variables

Create .env.local in the project root:

ANAKIN_API_KEY=your_anakin_api_key
OPENAI_API_KEY=your_openai_api_key
NEXT_PUBLIC_BASE_URL=http://localhost:3000

TURSO_DATABASE_URL=libsql://lathe-yourname.turso.io
TURSO_AUTH_TOKEN=your_turso_auth_token

Get your Anakin API key at anakin.io.

4. Seed the registry

Pre-populates the registry with Anakin's Wire catalog (50+ pre-built servers):

npm run seed

5. Run

npm run dev

Open http://localhost:3000.

How the Pipeline Works

Every URL goes through this pipeline:

Input URL
   │
   ├─ Wire catalog check (Anakin)
   │    HIT  → instant server, ~2 seconds ✓
   │    MISS ↓
   │
   ├─ Crawl site (Anakin /v1/crawl)
   │    ├─ Thin content? → retry with headless browser
   │    └─ Returns pages[] with markdown + generatedJson
   │
   ├─ Schema inference
   │    ├─ Anakin's data_schema → use directly (no AI cost)
   │    └─ Conflicting schemas → merge with gpt-4o-mini
   │
   └─ Generate MCP config + persist to Turso → Published ✓

Generation runs async. The UI polls /api/status/:jobId every 1.5s and shows live badge updates as each Anakin feature activates.

MCP Proxy

Every generated server is a real hosted MCP endpoint:

POST /api/mcp/:slug

Implements the MCP JSON-RPC 2.0 protocol (2024-11-05). Handles:

initialize — handshake with Claude Desktop
tools/list — returns one search tool per server
tools/call — fetches live data from Anakin (with 15-minute cache)

Wire-backed servers call the Anakin Wire action directly. Crawled servers re-crawl the source URL on demand.

Cache

Tool calls are cached for 15 minutes in Turso. Terminal output:

[MCP cache HIT]  hn-jobs — last crawled 2025-...   ← served from cache
[MCP cache MISS] hn-jobs — crawling Anakin now      ← hits Anakin

To force a re-crawl for a specific server:

UPDATE servers SET last_crawled_at = NULL, cached_data = NULL WHERE slug = 'hn-jobs';

Project Structure

app/
├── page.tsx                  # Homepage — URL input + live badge strip
├── registry/
│   ├── page.tsx              # Server gallery (Server Component)
│   └── ServerCard.tsx        # Registry card (Client Component)
├── server/[slug]/
│   ├── page.tsx              # Server detail + install instructions
│   └── CopyButton.tsx        # Copy to clipboard button
└── api/
    ├── generate/route.ts     # POST — starts generation pipeline
    ├── status/[jobId]/route.ts # GET — job polling
    └── mcp/[slug]/route.ts   # MCP JSON-RPC proxy endpoint

lib/
├── db.ts                     # Turso client + all queries
├── anakin.ts                 # Crawl, Wire action execution, Agentic Search
├── wire.ts                   # Wire catalog lookup
├── schema.ts                 # Schema inference (Anakin-first, OpenAI fallback)
├── utils.ts                  # timeAgo helper
├── types.ts                  # Shared TypeScript types
└── generators/
    └── mcp.ts                # MCP config generation, slug helpers

scripts/
└── seed-wire.ts              # Pre-seed registry from Wire catalog

API Reference

`POST /api/generate`

Start the generation pipeline.

// Request
{ "url": "https://news.ycombinator.com/jobs" }

// Response (202)
{ "jobId": "550e8400-e29b-41d4-a716-446655440000" }

`GET /api/status/:jobId`

Poll job progress.

{
  "id": "550e8400...",
  "status": "done",
  "step": "Published!",
  "badges": ["Map", "Scraping"],
  "server": {
    "slug": "news-ycombinator-com-jobs",
    "name": "News",
    "mcp_config": "{...}"
  },
  "error": null
}

Job statuses: pending → wire_check → crawling → inferring → generating → done / failed

`POST /api/mcp/:slug`

MCP JSON-RPC 2.0 endpoint. Used by Claude Desktop — not called directly.

`GET /api/mcp/:slug`

Returns server metadata (health check / discovery).

{
  "name": "Hacker News Jobs",
  "slug": "news-ycombinator-com-jobs",
  "mcp_endpoint": "https://lathe.dev/api/mcp/news-ycombinator-com-jobs",
  "protocol": "MCP JSON-RPC 2.0",
  "protocol_version": "2024-11-05"
}

Connecting to Claude Desktop

Generate a server at lathe.dev (or localhost:3000)
Copy the install snippet
Open ~/Library/Application Support/Claude/claude_desktop_config.json
Paste into the mcpServers object
Restart Claude Desktop

{
  "mcpServers": {
    "news-ycombinator-com-jobs": {
      "url": "https://lathe.dev/api/mcp/news-ycombinator-com-jobs"
    }
  }
}

Deploying to Vercel

vercel --prod

Set these environment variables in the Vercel dashboard:

Variable	Description
`ANAKIN_API_KEY`	From anakin.io
`OPENAI_API_KEY`	From platform.openai.com
`TURSO_DATABASE_URL`	From turso.tech
`TURSO_AUTH_TOKEN`	From turso.tech
`NEXT_PUBLIC_BASE_URL`	Your production URL, e.g. `https://lathe.dev`

After deploying, update NEXT_PUBLIC_BASE_URL and run the seed script once:

NEXT_PUBLIC_BASE_URL=https://your-app.vercel.app npm run seed

Anakin Features Used

Feature	Where	Anakin Endpoint
Wire catalog check	Before every crawl	`GET /v1/holocron/search`
Wire catalog seed	`npm run seed`	`GET /v1/holocron/catalog`
Wire action execution	MCP proxy live calls	`POST /v1/holocron/:id`
Multi-page crawl	Generation pipeline	`POST /v1/crawl`
Headless browser fallback	JS-heavy sites	`POST /v1/crawl` + `useBrowser: true`
Agentic Search	Topic input (no URL)	`POST /v1/agentic-search`

License

MIT

Name		Name	Last commit message	Last commit date
lathe		lathe
README.md		README.md

Navigation Menu

Repository files navigation

Lathe

What it does

Demo

Tech Stack

Getting Started

1. Clone and install

2. Set up Turso

3. Configure environment variables

4. Seed the registry

5. Run

How the Pipeline Works

MCP Proxy

Cache

Project Structure

API Reference

`POST /api/generate`

`GET /api/status/:jobId`

`POST /api/mcp/:slug`

`GET /api/mcp/:slug`

Connecting to Claude Desktop

Deploying to Vercel

Anakin Features Used

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

Lathe

What it does

Demo

Tech Stack

Getting Started

1. Clone and install

2. Set up Turso

3. Configure environment variables

4. Seed the registry

5. Run

How the Pipeline Works

MCP Proxy

Cache

Project Structure

API Reference

POST /api/generate

GET /api/status/:jobId

POST /api/mcp/:slug

GET /api/mcp/:slug

Connecting to Claude Desktop

Deploying to Vercel

Anakin Features Used

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

`POST /api/generate`

`GET /api/status/:jobId`

`POST /api/mcp/:slug`

`GET /api/mcp/:slug`