▸ WORKLOFT LABS · API · v0.1 · LIVE

Daily AI research picks your agent
can actually call.

The Workloft Labs pipeline now ships as an HTTP API and an MCP server. Five hand-scored arXiv picks per day, the published 9-axis rubric, REG FIT regulator-deployability. Free tier is live. No waitlist.

▸ Get a free key See what's on the bench →

5 hand-scored arXiv picks every morning, in your agent before breakfast.

REG FIT — the column nobody else publishes. Would this clear an FCA review?

Wire it once into any agent that speaks MCP. Forever.

▸ Substrate before spectacle ◂

▸ WHY THIS EXISTS

Most arXiv-watch tools answer "what's hot." This one answers "what survives a regulator." That's the column buyers in financial services, life sciences, and Local Authorities actually need.

The 9-axis rubric, published

A1–A9 covering substrate, governance, and research. Definitions are open. Scores are auditable. If you disagree with a pick, you can challenge the axis, not just the conclusion.

REG FIT

Every pick gets a 1–3 score for "would this clear an FCA Risk review or a UK GDPR DPIA?" — the column nobody else publishes and exactly the one regulated buyers need.

Built for agents, not dashboards

HTTP for any client; MCP server for Claude Code / Cursor / Cline. Wire labs_papers into your loop, never crawl arXiv yourself again.

Daily. Not weekly. Not "soon".

Walt screens 70+ arXiv + Hugging Face Daily papers every morning. Five clear the threshold. Citation-graphed via Semantic Scholar. Shipped before breakfast UK time.

▸ QUICKSTART · ONE MINUTE

curl, an email, and you're inside the loop.

1. Issue yourself a free key

curl -X POST https://chat-api.workloft.ai/labs-api/v1/keys \
  -H "Content-Type: application/json" \
  -d '{"email":"you@example.com","label":"my-first-agent"}'

2. Pull this week's picks

curl -H "X-API-Key: wl_labs_…" \
  "https://chat-api.workloft.ai/labs-api/v1/papers?days=7&min_reg_fit=2"

3. Wire the MCP server (optional)

// claude_desktop_config.json / cursor mcp settings
{
  "mcpServers": {
    "workloft-labs": {
      "command": "python3",
      "args": ["/path/to/labs_mcp.py"]
    }
  }
}

▸ MCP server source: labs_mcp.py — local-mode reads files directly, no HTTP roundtrip.

▸ ENDPOINTS

Base URL: https://chat-api.workloft.ai/labs-api · Auth: X-API-Key header

Method	Path	Auth	Description
GET	/v1/stats	—	pipeline-wide counts (days tracked, total picks)
GET	/v1/axes	—	the published 9-axis rubric
GET	/v1/papers	key	recent picks. Filters: `days`, `axis`, `min_score`, `min_reg_fit`, `limit`
GET	/v1/papers/{arxiv_id}	key	single pick by arXiv ID
GET	/v1/me	key	your usage, tier, and limits
POST	/v1/keys	—	self-issue a free-tier key (one per IP per 24h)

▸ Interactive docs at /v1/docs · OpenAPI JSON at /v1/openapi.json.

▸ TIERS

Free tier ships today. Paid tiers wire in week 3 once we see real traffic shape.

Tier	Calls / 30d	History	Max limit/call	Price
Free	100	7 days	25	£0
Pro soon	5,000	90 days	100	£49/mo
Team soon	50,000	1 year	500	£499/mo
Enterprise contact	unlimited	full	5,000	£5k+/mo
Bench-as-a-Service contact	Private Lab Bench tracking your firm's named papers, custom rubric, weekly Note			£15k/mo

Need a tier today? alfred@workloft.ai — we'll cut you a key by hand.

▸ MCP SERVER

Workloft Labs is an MCP server. Wire it once, every model in your stack reads the same picks.

Tools exposed:

▸ labs_papers — list recent picks with full filters
▸ labs_paper — fetch a single pick by arXiv ID
▸ labs_axes — the 9-axis rubric
▸ labs_stats — pipeline-wide counts

Local-mode (file reads, offline-capable) is open source. Hosted MCP fronting the HTTP API is in week-2 work — emails to alfred@workloft.ai for early access.

Daily AI research picks your agentcan actually call.