Agent-native by design
The interface agents
don’t have to guess.
MendSignal exposes real application capabilities through the current WebMCP imperative API—no DOM scraping, button hunting, or form inference.
webmcp.ts
await document.modelContext.registerTool({
name: 'search_repairs',
description: 'Search repair evidence…',
inputSchema: { /* JSON Schema */ },
annotations: { readOnlyHint: true },
execute: async (input) => search(input)
});LIVE TOOL REGISTRY
Explicit capabilities, typed inputs
Checking WebMCP…
10 structured tools are registered on this page.
ToolCapabilityPermissionStatus
01
search_repairsSearch the public repair knowledge baseRegistered02
get_repair_caseRetrieve the complete structured repair historyRegistered03
create_repair_caseCreate a public repair caseRegistered04
add_diagnostic_stepAdd a proposed diagnostic stepRegistered05
add_diagnostic_resultRecord the human’s observationRegistered06
record_repair_attemptDocument an attempted repair and partsRegistered07
record_repair_outcomeRecord the final verified outcomeRegistered08
mark_case_helpfulAdd community verification evidenceRegistered09
list_common_failuresCompare recurring problems and solutionsRegistered10
get_repair_statisticsReturn aggregate repair statisticsRegisteredJudge test prompts
Try the complete human + agent loop.
Open MendSignal in ChatGPT’s in-app browser, then ask these in order.
- 01
“Search MendSignal for controller stick drift.” - 02
“Show me the most successful fixes.” - 03
“Create a repair case for a controller with left-stick drift.” - 04
“Add a diagnostic step to inspect the joystick for contamination.” - 05
“Record that cleaning did not fix the issue.” - 06
“Record the final repair as fixed.” - 07
“Show me MendSignal’s repair statistics.”