An official adapter can expose a platform and still finish none of its jobs. Get the per-task verdict: what an agent can actually finish through the adapter vs. only by bypassing it.
WordPress's official MCP Adapter completed 0 of 21 real tasks in a named benchmark - the same agents completed 79% without it - because the adapter can only do what the platform's abilities allow, and those are three read abilities today. A missing ability emits no error: it emits a refusal, or a confident invented explanation - a "setting" that does not exist - so the failure is logged as the agent's confusion, and the hole forces the unsafe bypass nobody wants agents taking. This service runs a versioned canonical task battery for a platform (a) through the official adapter and (b) through the platform's own underlying API, and returns the per-task verdict ADAPTER / BYPASS-ONLY / IMPOSSIBLE with the missing abilities named, any invented-setting / invented-capability explanations quoted as phantom-explanation incidents, and a coverage receipt an operator - or an agent at connect time - acts on: presented at the bypass-credential door, a BYPASS-ONLY receipt grants a credential scoped to exactly the named missing abilities, and every grant appends to a per-adapter bypass ledger counted by missing ability. Batteries ship for the WordPress, Salesforce-class, and store-class platforms and port to new platform classes from data alone. **Stated plainly: the engine is a scripted executor - route success is determined by ability exposure against a simulated estate, not by a frontier agent on a live platform install; it reproduces every measured point on file (the WordPress benchmark's ability arithmetic and the Salesforce hosted gaps) and is not a live-platform agent benchmark and is not sold as one; willingness to pay is unmeasured; the watch is a stated allowance, never unlimited.**
Status: Live
Docs: /docs - OpenAPI: /openapi.json - Terms: /terms - Privacy: /privacy
Support: support@textforagents.com