Use Rhumb through the Model Context Protocol — the fastest path to agent tool discovery and execution.
For most agents, the default production path is:
- discover a Service or Capability
- resolve ranked providers, or follow the machine-readable recovery handoff
- estimate the call
- execute through Layer 2 with
RHUMB_API_KEY
Use x402 only when zero-signup per-call payment is the point. For repeat traffic, use governed API key or wallet-prefund on X-Rhumb-Key, and bring BYOK or Agent Vault only when provider control is the point.
npx rhumb-mcp@latestZero config. Discovery tools work immediately.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"rhumb": {
"command": "npx",
"args": ["rhumb-mcp@latest"],
"env": {
"RHUMB_API_KEY": "your_key_here"
}
}
}
}The governed API key is optional — discovery tools work without it. Add it to enable execution.
In Cursor settings → MCP Servers → Add:
{
"rhumb": {
"command": "npx",
"args": ["rhumb-mcp@latest"],
"env": {
"RHUMB_API_KEY": "your_key_here"
}
}
}find_services— "Find the best email API for agents"get_score— "What's Stripe's agent-native score?"get_alternatives— "What are alternatives to SendGrid?"get_failure_modes— "What breaks when agents use HubSpot?"discover_capabilities— "What capabilities exist for payments?"
resolve_capability— "Which provider or recovery handoff should I use for email.send, and what if I only want BYOK or Rhumb-managed?"estimate_capability— "What execution rail, health, and cost should I expect before this call runs?"execute_capability— "Send an email through the best provider"get_receipt— "Show me the receipt for that execution"usage_telemetry— "How are my calls performing this week?"budget/spend— "Set a $10/day budget"
discover_capabilities— find the action you wantresolve_capability— see ranked providers, optional credential-mode filters, machine-readable recovery fields likerecovery_hint.resolve_url,recovery_hint.credential_modes_url, and, when applicable,recovery_hint.alternate_execute_hintorrecovery_hint.setup_handoff, plus search suggestions when the capability ID is wrongestimate_capability— check the active execution rail, health, and cost before executionexecute_capability— run the action
If you ask for a specific mode such as byok or rhumb_managed and nothing is execute-ready, resolve_capability can still return the next action as machine-readable recovery_hint.alternate_execute_hint or recovery_hint.setup_handoff instead of leaving you to rediscover it manually. If you typo the capability ID, it can also return a capability search URL plus suggested capabilities instead of a blank dead end.
- Go to rhumb.dev/auth/login
- Sign in with GitHub or Google
- Copy your governed API key from the dashboard
You: Find me the best API for sending transactional emails from an AI agent.
Agent (using find_services): Found 8 email services. Top 3 by AN Score:
- Resend: 7.8 (L3 Ready) — cleanest API, best agent ergonomics
- Postmark: 6.8 (L3 Ready) — reliable delivery, some signup friction
- SendGrid: 5.5 (L2 Developing) — powerful but complex auth model
You: What are Resend's failure modes?
Agent (using get_failure_modes): 3 known failure modes for Resend:
- Domain verification required before sending (blocks cold start)
- Rate limit headers not always present
- Webhook signatures use asymmetric keys (unusual pattern)
You: Ok, send a test email through Resend.
Agent (using execute_capability): Executing email.send via Resend...
- Cost estimate: $0.001
- Status: 200 OK
- Message ID: msg_abc123
- Discovery works with no key.
- Execution through governed API key or wallet-prefund uses
RHUMB_API_KEY(your governed API key). - Use x402 through the REST API when zero-signup per-call payment matters (no
RHUMB_API_KEYrequired). - Use wallet-prefund when the same wallet needs repeat traffic.