surface: homepage purpose: choose the correct Rhumb route first_action: search Index index_definition: discovery and comparison resolve_definition: governed execution when callable before_execution: inspect AN Score, evidence, freshness, credential mode, and estimate if_not_callable: do not invent execution; return best next step machine_entrypoints: MCP, REST API, /llms.txt, /.well-known/agent-capabilities.json current_public_truth: Discovery breadth: 999 scored services and 435 capability definitions. Current runtime-callable surface: 18 callable providers, strongest today for research, extraction, generation, and narrow enrichment.

THE NAVIGATION LAYER FOR AGENT TOOLS

From task to trusted tool call.

Rhumb is the tool discovery and governed access layer for AI agents: Index shows what can be trusted, and Resolve runs what is actually supported.

Discovery leads. Execution follows only when a route is callable. If coverage is missing, Rhumb returns the next best step instead of pretending.

Run the MCP server Browse the Index

Broad discovery first. Governed execution when ready. Honest stop when unsupported.

llms.txt agent-capabilities.json docs 21 MCP tools
Read-only route trace Index → Evidence → Resolve → Stop
  1. 01 · Task Intent received

    Research Acme and summarize the current facts.

    A human gives the agent an outcome, not a provider decision.

  2. 02 · Index Compare routes

    Search capability: company.research

    Index is discovery and comparison: services, capabilities, AN Score, evidence, freshness, and failure modes.

  3. 03 · Evidence AN 91 · Fresh

    Inspect trust before action.

    Check score, proof, freshness, credentials, estimate, and failure modes before the agent acts.

  4. 04 · Resolve 420ms · $0.02

    Execute only when supported.

    Resolve is governed execution for supported capabilities: estimate, apply policy, route, and run.

  5. 05 · Outcome Execute / stop

    Return the result or stop honestly.

    If execution is unsupported, do not invent it. Return the evidence and next best step.

Stop rule active If unsupported, return best next step.
999+ services scored
435 capabilities mapped
18 callable providers
API + MCP governed surfaces

Dual-audience surface

Humans need confidence. Agents need procedure.

Rhumb is designed so a person can trust the route, and an agent can operate it without guessing.

Start here.

For humans

  1. Understand what’s possible Search the Index by task, capability, or provider.
  2. Choose with confidence Use AN Score, evidence, freshness, and failure modes.
  3. Execute when ready Use Resolve when Rhumb shows a callable path.
See the human path →

Point your agent here.

For agents

Machine docs

Starter route for autonomous agents

Paste this into your agent to send it down the Rhumb operating path.

surface: homepage
purpose: choose the correct Rhumb route
first_action: search Index
use_resolve_when: capability is callable
before_execution:
 - inspect AN Score
 - inspect evidence and freshness
 - inspect credential mode and estimate
if_not_callable:
 - do not invent execution
 - return best next step
machine_entrypoints:
 - MCP
 - REST API
View full agent guide →

Reliable bearing

Two journeys. One reliable bearing.

Rhumb helps both the human and the agent become more capable.

Human journey

Problem

My agent needs tools, but the landscape is risky.

Guide

Rhumb shows the safest route.

Action

I launch my agent with Index or Resolve.

Reward

I get a more capable, trustworthy agent.

Agent journey

Problem

I have a task but no reliable route.

Guide

Rhumb maps capabilities and evidence.

Action

I inspect, decide, and execute or stop honestly.

Reward

I become the hero for my human.

Product model

Find. Trust. Execute.

Use the right Rhumb surface at the right time.

01 · Index

Find with Index

Search the Index for capabilities by task or provider.

company.researchweb.searchmath.solve
Explore Index →

03 · Resolve

Execute with Resolve

Execute only when the route is callable and governed.

Try Resolve →

Evidence ledger

Why agents can explain the route.

Every good call has a source trail, a freshness boundary, and a visible stop condition.

Evidence tiers Clear quality signals you can trust.
Freshness Know how current the data is.
Failure modes See what can go wrong.
Dispute path Open issues. We listen and improve.
Evidence ledger sample
SourceTypeFreshnessAN Score
Acme Investor RelationsWeb2h ago95
CrunchbaseAPI5h ago90
Company WebsiteWeb6h ago88
News APIAPI12h ago82

Quickstart

Start building in minutes.

MCP, REST, or a starter prompt. The MCP path is the fastest way to give an agent a reliable bearing.

REST Starter prompt
terminal 
$ npx rhumb-mcp
> starting Rhumb MCP server...
> connected
> tools: find_services, get_score, get_failure_modes, resolve_execute
> find_services("company research")
> get_score("Acme API", "company.research")
> resolve_execute("best_route", "company.research")

Hold the heading

Give your agent a reliable bearing.

Run MCP Browse free View docs