MAPS + APPS · Capability Navigator
WebMCP reference · 2026

Human + agent, same evidence

Agents shouldn’t have to guess what your website means.Give them tools, not pixels.

Capability Navigator turns a fuzzy software need into an evidence-backed, nonbinding decision packet. A human and browser agent work from the same public capability records and the same visible review surface.

Checking WebMCP…human fallback loads either way
list_capabilitiesget_capabilityfind_examplesdraft_scopeprepare_decision_packetstage_human_review
6 tools

The live handoff

One workspace. Two operators.

Build the packet yourself, or ask a WebMCP-aware agent to call stage_human_review. Either way, the result lands in the same visible interface for human inspection and editing.

Demo path · ~60 sec

Ask the page this.

Using only this page's WebMCP tools, help me scope a public-data operational dashboard. Find the relevant capabilities and inspect the public evidence. Then stage a human-review packet in the visible page using the relevant capability IDs, with these constraints: Public data only; Mobile-first. Do not contact anyone, submit anything, pay, sign, commit, or perform any external action.
01Intent
02Typed tools
03Public evidence
04Human review
Human and agent share one evidence model
Evidence-backed capabilities

Agent trace

Every WebMCP invocation is reflected visibly here. The page does not hide agent activity behind a server log.

    Tool manifest

    Six narrow tools. Human authority stays visible.

    list_capabilitiesDiscover relevant public capabilities.READ ONLY
    get_capabilityInspect one capability and its proof boundary.READ ONLY
    find_examplesRetrieve linked public evidence.READ ONLY
    draft_scopeCreate a deterministic nonbinding scope draft.READ ONLY
    prepare_decision_packetBind scope, evidence, questions, and exclusions.READ ONLY
    stage_human_reviewStage the packet in current-tab UI for a person to inspect.LOCAL UI ONLY