Migration Guide

Switching from Smithery to Rhumb

Smithery and Rhumb solve related but different problems. Smithery is the tool shelf — the largest MCP server catalog available. Rhumb is the quality layer — trust scores, failure documentation, and agent-native execution. Many teams use both.

This guide is honest about where Smithery wins. We believe the best way to earn trust is to tell you exactly when a competitor is the better choice.

TL;DR

  • → Use Smithery if you need the widest catalog, visual browsing, or quick experimentation with new MCP servers.
  • → Use Rhumb if you need to know which tools actually work for agents, how they fail, what they cost per-call, and you want autonomous agent payments.
  • → Use both if you discover on Smithery and validate on Rhumb before production deployment.

Head-to-head comparison

10 dimensions. We call it honestly — including where Smithery wins.

Catalog size

Smithery

Smithery

4,000+ MCP servers indexed

Rhumb

999 services scored across 92 categories

Quality signal

Rhumb

Smithery

Star ratings, install counts

Rhumb

20-dimension AN Score: execution, access, autonomy. Machine-readable. Confidence-weighted.

Failure documentation

Rhumb

Smithery

Not available — agents discover failures at runtime

Rhumb

Failure modes documented per service. Agents know how a tool fails before calling it.

Credential paths

Rhumb

Smithery

Managed OAuth (human-session-oriented)

Rhumb

Three credential paths: Rhumb-managed (Rhumb holds the provider credential), BYOK (you keep the provider API key), Agent Vault (Rhumb injects your encrypted provider credential at call time). x402 stays separate as the zero-signup payment rail.

Agent payments

Rhumb

Smithery

Credit card billing, subscription tiers

Rhumb

x402 per-call USDC on Base — agents pay at the HTTP layer, no signup required

Free tier

Smithery

Smithery

25,000 RPCs/month

Rhumb

Discovery is always free. Executions are pay-as-you-go with route, health, and cost checks before execution.

Onboarding speed

Smithery

Smithery

Web UI, one-click install, visual browser

Rhumb

API-first with curl examples. MCP via npx. No web UI for browsing yet.

Cost transparency

Rhumb

Smithery

Per-plan pricing ($0/free, $30/pro)

Rhumb

Per-call route, health, and cost checks before execution. You see the exact path and price before committing.

Alternatives / routing

Rhumb

Smithery

Manual browsing of similar servers

Rhumb

460+ alternative relationships mapped. API returns scored alternatives automatically.

Protocol support

Tie

Smithery

MCP-native, hosted MCP servers

Rhumb

MCP + REST API. 16 MCP tools. Full REST coverage for non-MCP agents.

Score: Smithery wins on 3 dimensions (catalog, free tier, onboarding). Rhumb wins on 6 (quality signal, failure docs, credentials, payments, cost transparency, alternatives). 1 tie (protocol support). The pattern is clear: Smithery wins breadth and ease; Rhumb wins depth and trust.

When to stay with Smithery

You're exploring, not deploying

Smithery's visual browser and 4,000+ server catalog is unbeatable for discovery and experimentation. If you're trying tools to see what's possible, start there.

You need a niche MCP server

With 8x our catalog size, Smithery likely has tools we haven't scored yet. If the tool you need isn't in Rhumb's directory, Smithery probably has it.

Your budget is zero

Smithery's free tier (25,000 RPCs/month) is generous. Rhumb charges per call from the first call, though discovery is always free.

When to switch to Rhumb

Your agent needs to choose tools autonomously

Star ratings and install counts don't tell an agent whether a tool's auth flow works without human intervention. The AN Score measures exactly that — execution quality, access readiness, and autonomy dimensions that matter for unattended operation.

You've been burned by silent failures

Rhumb documents how every scored service fails — before your agent encounters it. Failure modes include blast radius, recovery steps, and prevention guidance. Your agent knows the failure surface of every tool before making a single call.

Your agent needs to pay for itself

x402 payments let agents pay per-call with USDC at the HTTP layer when zero-signup, request-level payment authorization is the point. It is powerful, but it is not the default repeat-traffic rail; for that, use governed API key or wallet-prefund on X-Rhumb-Key, and bring BYOK or Agent Vault only when provider control is the point.

You need production reliability guarantees

Rhumb's fail-closed billing, per-agent rate limiting, budget controls, and documented runbooks are built for production. If an upstream provider is degraded, you get a 502 with an alternatives endpoint — not a cryptic timeout.

Migration steps

If you're adding Rhumb alongside or replacing Smithery, here's the path.

1

Audit your current tool usage

List the MCP servers you're actively calling through Smithery. Search for each one on Rhumb: curl https://api.rhumb.dev/v1/search?q=resend

2

Check AN Scores for your tools

For each service, review the AN Score, failure modes, and alternatives: curl https://api.rhumb.dev/v1/services/resend/score

3

Choose your access path

Start with Rhumb-managed if you want us to handle credentials. Bring BYOK if you already have provider API keys. Use Agent Vault if you want Rhumb to inject your encrypted provider credential at call time. Use x402 only when zero-signup per-call payment is the point. Quickstart guide →

4

Connect via MCP or REST

MCP: npx rhumb-mcp — drop-in alongside your Smithery connection. REST: call https://api.rhumb.dev/v1 directly.

5

Run both in parallel

There's no reason to hard-cut. Use Smithery for discovery and experimentation. Use Rhumb for production execution and trust decisions. They're complementary.

What Rhumb doesn't do (yet)

  • • Visual web UI for browsing tools (API and MCP only today)
  • • Hosted MCP servers (Rhumb is a layer, not a host)
  • • 4,000+ server catalog (we're at 999 scored services and growing daily)
  • • Team management and role-based access (planned)
  • • Usage analytics dashboard (planned)
Next honest step

Move from catalog breadth to one governed execution lane

If discovery is no longer the blocker, the next honest move is to bound execution, choose the right credential path, and start with one governed lane instead of a bigger shelf. Use capability-first onboarding when you still need to sort trust and workflow fit, or open the managed path directly when you already know the bounded job you want Rhumb to run.

See the capability-first handoff → Open the managed path →
Fleet follow-through

Discovery breadth is only useful if the overnight lane stays bounded

Once catalog choice is settled, the production questions become credential lifecycle and shared-budget control. These guides take the conversation from tool discovery into the two fleet layers that usually break next.

API Credentials in Autonomous Agent Fleets

Maps the real production work after a tool is chosen: rotation, revocation, expiry handling, and shared-key cascade prevention.
Designing Agent Fleets That Survive Rate Limits

Shows why catalog quality still is not enough when bursts, retries, and quota sharing are left ungoverned after launch.