open standard · io.glimind/reliability

The reliability standard for AI agent tools

MCP tells an agent a tool exists and how to call it — but there is no standard way to say whether it's working right now, or to carry a neutral reliability rating. Glimind publishes one: an open, payload-free schema any server, registry, gateway, or agent can emit, read, and verify.

Three parts

1 · A JSON Schema

A canonical reliability report shape (verdict, recommendation, score/grade, call hints, expected output, forecast, schema-drift, setup). Draft 2020-12.

schema.json →

2 · An MCP extension

First-class 2026-07 capability io.glimind/reliability advertised in server/discover — points consumers at health + the verification keys.

descriptor →

3 · Signed attestations

Every rating + the feed is an Ed25519 compact-JWS, verifiable against our JWKS — so a Glimind rating is tamper-evident and provably neutral.

jwks.json →

Why a neutral standard

Reliability data is only trustworthy if it's neutral and verifiable. Glimind sells no tools and runs no integrations, so its rating is unconflicted; signatures make it tamper-evident. The schema is open and free to implement — adopt it to expose health uniformly, or just consume Glimind's signed ratings.

Verdicts & recommendations

verdictrecommendationmeaning
healthyproceedworking now; standard retries
degradedproceed_with_cautionelevated failures; line up an alternative
downavoidbroken now; route away
unknownunknowninsufficient signal yet

Use it

# discover the standard
curl https://glimind.com/.well-known/reliability.json
# a signed, schema-conformant report for a tool
curl https://glimind.com/v1/score/github%2Fsearch_repositories
# verify: the report's attestation.jws is an Ed25519 JWS over its payload
curl https://glimind.com/.well-known/jwks.json

Proposing this to the MCP community as an SEP — see the draft in the repo (launch/SEP-reliability.md). Feedback welcome. Consume it today via any integration.