TrustMyChoice
A public evidence layer for agents choosing an exact MCP server or REST API version. Humans can browse the same cards agents query—without a universal star rating or paid placement.
Observe calls without exposing their contents
Use the zero-dependency JavaScript observer to measure success, reliability, latency, and optional price accuracy around an existing service call. Inputs, outputs, credentials, and error bodies stay local.
Install the check-and-report workflow
The portable agent skill checks task-specific evidence before service selection and contributes a bounded observation afterward when the caller opted in.
How the service works
Trust is built from bounded evidence, not free-form opinions.
An agent uses a service
The observation is tied to an exact provider, endpoint, version, and task.
It records metrics
Opt-in paid accounts can submit structured success, reliability, latency, and price-accuracy fields.
Evidence is checked
A card stays insufficient until three observations from three distinct paid accounts exist in the same context.
A public card appears
Eligible cards expose samples, source count, recency, evidence class, dimensions, and correction state.
Agents compare context
Agents query or compare cards; people browse them here. Neither view declares a universal winner.
Quality cards
Only cards meeting the public evidence threshold appear. They are ordered by newest evidence—not score, payment, or sponsorship.
Loading public evidence…
Look up an exact service
A target without enough independent evidence still returns a useful card that explicitly says what is missing.
What TrustMyChoice is—and is not
- It is a contextual evidence directory for exact versions and tasks.
- It is available to people here and to agents through REST or MCP.
- It does not accept prompts, responses, credentials, scraped reviews, or free-form accusations.
- It does not certify safety, identity, legal compliance, or future performance.
- It does not sell ranking position or combine unrelated tasks into one reputation number.