Pay-per-call ($0.01 USDC) model-routing for AI coding agents: which LLM to call, cost vs. quality.
https://routing-courier-mcp-production.up.railway.app/mcpanswers HTTP, not MCP
| Last 30 probes | |
|---|---|
| Last seen healthy | 2026-10-18 |
| Completed sessions | 28.6% of the last 7 probes · 82.8% of the last 30 |
| Handshake latency | median 163 ms, 95th 198 ms |
| Tools listed | — |
| Tool contract | 7bf57a71f31c — unchanged over the 18 days we have read it |
| Protocol answered | — |
| Transport | streamable-http |
| Last detail | HTTP 404 |
[](https://referencesource.org/mcp-health/server/io-github-carlos-vlab-routing-services-courier/)
The badge shows the most recent probe. It is generated from the same file this page is, updates once a day, and needs no key.
One HTTP POST carrying a JSON-RPC initialize, and if
that succeeds, a tools/list. Nothing here is a score, a rating or a
judgement about whether the server is any good — only whether an ordinary MCP
client could complete a session with it on the day we asked.
“Needs credentials” is not a fault: it means the endpoint wants
an account, which is a perfectly reasonable thing for it to want.
The tool contract row is a digest of what
tools/list returned — each tool's name, description and input
schema, hashed — so that a rename, a newly-required field or a rewritten
description can be spotted even though none of them breaks the connection. We
keep the digests and nothing else: no tool name, description or schema of
anybody's is stored or republished here, which is also why this can say that a
contract moved and by how much, but not what it now says.
The method and the daily totals · every server we probe, A–Z · feed.json · licence · corrections and removals: [email protected]