Rhumb

Agent-native tool intelligence. Discover, score, and execute APIs — with cost-aware routing, budget enforcement, and three credential modes. Built for autonomous agents, not humans browsing docs.

🌐 rhumb.dev · 📊 Leaderboard · 📖 Methodology · 🔌 MCP Server · 🔑 Trust

What is Rhumb?

Rhumb is the infrastructure layer agents use to discover, access, and trust external tools.

• 212 scored services across 38 categories, evaluated on 20 dimensions
• 103 capabilities across 30 domains (email.send, project.create_issue, ai.generate_text, spreadsheet.read, ...)
• 3 credential modes — bring your own token, zero-config managed, or self-provisioned via ceremony guides
• Cost-optimal routing — cheapest provider that meets your quality floor
• Budget enforcement — pre-execution checks, not surprise bills

MCP Server

npx rhumb-mcp@0.5.0

13 tools:

Tool	What it does
find_tools	Search services by category or capability
get_score	AN Score and dimension breakdown for a service
get_alternatives	Find alternatives ranked by score
get_failure_modes	Known failure modes and workarounds
discover_capabilities	Browse capabilities by domain
resolve_capability	Ranked providers with health + cost for a capability
execute_capability	Execute through Rhumb — auto-selects provider, injects auth
estimate_capability	Cost estimate + budget check without executing
credential_ceremony	Step-by-step guide to get API credentials for a service
check_credentials	Your credential status across all three modes
budget	Check or set your execution budget
spend	Spend breakdown by capability and provider
routing	Get or set your routing strategy

Quick Example

# "I need to send an email. What should I use?"
resolve_capability("email.send")
→ 8 providers ranked: Resend (8.3), SendGrid (7.5), Postmark (7.2), ...

# "Use the cheapest one that's good enough"
routing(set, strategy="cheapest", quality_floor=7.0)

# "Send it — zero config, Rhumb handles credentials"
execute_capability("email.send", credential_mode="rhumb_managed", body={
  from: "agent@example.com", to: "user@example.com",
  subject: "Hello", html: "<p>Sent via Rhumb</p>"
})
→ { provider_used: "resend", upstream_status: 200, budget_remaining_usd: 49.99 }

Three Credential Modes

Mode	How it works	When to use
BYO Token	You provide your own API key	Full control, existing credentials
Rhumb Managed	Zero-config — Rhumb provides credentials	Quick start, managed capabilities
Agent Vault	Self-provision via ceremony guide, pass per-request	Your own key, Rhumb never stores it

AN Score

The Agent-Native Score rates services from 0–10 across two axes:

Axis	Weight	What it measures
Execution	70%	API reliability, error ergonomics, schema stability, latency, idempotency, autonomy
Access Readiness	30%	Signup friction, credential management, rate limit transparency, docs quality, sandbox availability

Tiers:

• L4 Native (8.0–10.0) — Built for agents
• L3 Ready (6.0–7.9) — Works reliably with minor friction
• L2 Developing (4.0–5.9) — Usable with workarounds
• L1 Emerging (0.0–3.9) — Significant barriers to agent use

Full methodology: rhumb.dev/methodology

REST API

Base URL: https://rhumb-api-production-f173.up.railway.app/v1

Endpoint	Description
GET /capabilities	Browse capability registry
GET /capabilities/{id}/resolve	Ranked providers for a capability
POST /capabilities/{id}/execute	Execute a capability
GET /capabilities/{id}/execute/estimate	Cost estimate
GET /capabilities/rhumb-managed	Zero-config managed capabilities
GET /services/ceremonies	All credential ceremony guides
GET /services/{slug}/ceremony	Ceremony steps for a specific service
GET/PUT /agent/budget	Budget management
GET/PUT /agent/routing-strategy	Routing strategy
GET /agent/spend	Spend breakdown
GET /services/{slug}	Service details + score
GET /search?q=...	Service search

Project Structure

rhumb/
├── packages/
│   ├── web/      # Astro 6.0 frontend (rhumb.dev)
│   ├── mcp/      # MCP server (npx rhumb-mcp)
│   ├── api/      # FastAPI on Railway (Supabase REST)
│   ├── cli/      # CLI tools
│   └── shared/   # Shared types and constants
├── scripts/      # Scoring and verification scripts
└── artifacts/    # Dataset and score artifacts

Development

# API
cd packages/api && pip install -r requirements.txt && uvicorn app:app --reload

# MCP
cd packages/mcp && npm ci && npm run dev

# Web
cd packages/web && npm ci && npm run dev

# Tests
cd packages/api && python -m pytest        # 80+ API tests
cd packages/mcp && npm test                 # 66 MCP tests

Score Disputes

Every score is disputable. If you believe a score is inaccurate:

1. GitHub Issue: Open an issue with evidence
2. Email: providers@supertrained.ai

All disputes are reviewed and outcomes are public.

Links

• Website: rhumb.dev
• About: rhumb.dev/about
• Trust: rhumb.dev/trust
• Blog: rhumb.dev/blog
• Twitter: @pedrorhumb

License

MIT