bullet-heaven.ahraz.me WebMCP audit

Name: bullet-heaven.ahraz.me WebMCP audit
Creator: ora

Rescan

audited Aug 31, 2026in 20s

81/ 100100% of the applicable check weight evaluated

20 tools registered. Strongest in shared experience, weakest in task completion.

Shared experience100

Task completion60

Tool quality83

Trust75

https://bullet-heaven.ahraz.me

Tools

4 read, 16 write

Loading map

get_maker_state

Returns { project, playtest }: the versioned project document and the current playtest session. Read it when you start work, when you need wave, emitter or pattern ids you do not already have, and again if the person may have edited since your last call, because human and agent edits share one live document. You do not need to call it before or after a mutating tool: each one returns the state it changed.Answers a question. Declared read only.

pagehttps://bullet-heaven.ahraz.me/

implementation

viaimperative

entry pointdocument

registered after196ms

executepresent

api surfaceTakes no parameters.

annotations

read onlytrue

untrusted contentnot declared

titlenot declared

1 finding

warningInjection surface

weight 4

14 of 20 tools carry instruction-shaped text in their own metadata: get_maker_state Description instructs when to call and when not to call the tool.

get_maker_state: Description instructs when to call and when not to call the tool; update_arena: Description instructs to call another tool before reporting or editing; update_player: Description instructs to call another tool before reporting or editing; open_editor: Description instructs when to call and what to check before calling

fix

{
  name: "get_maker_state",
  // Metadata DESCRIBES the tool - it never addresses the agent reading it.
  // Rewrite any name, description, title, or schema field description that
  // instructs the agent (which tool to prefer, what to output, rules to
  // ignore) so it states what the tool does and what it returns instead.
  description: "Searches the catalog and returns matching items with prices."
}

tool json

{
  "name": "get_maker_state",
  "description": "Returns { project, playtest }: the versioned project document and the current playtest session. Read it when you start work, when you need wave, emitter or pattern ids you do not already have, and again if the person may have edited since your last call, because human and agent edits share one live document. You do not need to call it before or after a mutating tool: each one returns the state it changed.",
  "inputSchema": {
    "type": "object",
    "properties": {},
    "additionalProperties": false
  },
  "annotations": {
    "readOnlyHint": true
  }
}

Findings

Shared experience

100 / 100 · weight 30

passPage experience

weight 20

The page a person sees holds up next to the agent surface: a working interface, visible actions, readable content, nothing in the way.

passHuman parity

weight 10

The person co-browsing can see and use this page - the same page the agent's tools act on.

Task completion

60 / 100 · weight 25

failTool selection

weight 15

An agent chose an existing tool and built a schema-valid call that held up on 1 of 3 canonical intents; on "Open the document I was working on most recently." no tool on the page served it. Nothing was executed, so this verifies selection, not outcomes.

passCoverage vs. site type

weight 10

The tool set covers the whole core editor flow: open or read a document, edit or write content, save, publish, or share the result.

Tool quality

83 / 100 · weight 25

passRegistration timing

weight 1

Every measured tool registered within 197ms of navigation.

tooledit_firing_pattern

passCanonical entry point

weight 3

All 20 tools register on the canonical document.modelContext entry point.

warningSchema validity

weight 4

1 of 20 declared input schemas cannot be read by an agent: apply_encounter input schema nests deeper than 6 levels at inputSchema.waves.items[0].emitters.items[0].patterns.items[0].name.

toolapply_encounter

apply_encounter: input schema nests deeper than 6 levels at inputSchema.waves.items[0].emitters.items[0].patterns.items[0].name

fix

{
  name: "apply_encounter",
  inputSchema: { type: "object", properties: { /* ... */ }, required: [] }
}

warningSchema quality

weight 4

18 of 20 tool schemas are harder for an agent to use than they need to be: evaluate_encounter description is 1896 characters, over Chrome's 500-character guidance.

toolevaluate_encounter

evaluate_encounter: description is 1896 characters, over Chrome's 500-character guidance; evaluate_encounter: 1 parameter description is over Chrome's 150-character guidance; update_arena: description

fix

{
  name: "evaluate_encounter",
  inputSchema: {
    type: "object",
    properties: {
      query: { type: "string", description: "What to search for" }
    },
    required: ["query"]
  }
}

warningNaming quality

weight 2

Tool naming makes selection harder than it needs to be: 7 names do not start with a verb (evaluate_encounter).

toolevaluate_encounter

fix

{ name: "evaluate_encounter" /* short, unique, verb-based */ }

passStub detection

weight 4

All 20 tools declare an execute handler.

passRegistration errors

weight 2

No tool registration threw during the capture.

passDescription quality

weight 5

All 20 rated tool descriptions say what the tool does, when to use it, and what it returns.

Trust

75 / 100 · weight 20

passAnnotations present

weight 6

All 4 read-shaped tools declare readOnlyHint: true - the one declared claim that lets an agent treat a call as safe to make without asking.

passAnnotation mismatch

weight 6

No tool claims to be read-only while its own name or description says it writes (20 declared hints checked).

not applicableUntrusted content hint

weight 4

No tool's contract suggests it returns text written by somebody other than the site, so there is nothing to flag as untrusted.

warningInjection surface

weight 4

14 of 20 tools carry instruction-shaped text in their own metadata: get_maker_state Description instructs when to call and when not to call the tool.

toolget_maker_state

get_maker_state: Description instructs when to call and when not to call the tool; update_arena: Description instructs to call another tool before reporting or editing; update_player: Description instructs to call another tool before reporting or editing; open_editor: Description instructs when to call and what to check before calling

fix

{
  name: "get_maker_state",
  // Metadata DESCRIBES the tool - it never addresses the agent reading it.
  // Rewrite any name, description, title, or schema field description that
  // instructs the agent (which tool to prefer, what to output, rules to
  // ignore) so it states what the tool does and what it returns instead.
  description: "Searches the catalog and returns matching items with prices."
}

Tool selection

33% across 3 intents

Open the document I was working on most recently.missed

choseno toolexpected document openkind not applicable

No tool was picked for this intent.

arguments

{}

Add a short summary paragraph at the top.missed

choseno toolexpected content editkind not applicable

No tool was picked for this intent.

arguments

{}

Save it and give me a link I can share.ok

chosesave_projectexpected save or sharekind matched

arguments

{}

model: gemini-2.5-flash

Add the tools this site is missing

Our scanner reads your website and suggests the right WebMCP tools for it.

The open source webmcp plugin teaches your coding agent to audit a site, implement tools on document.modelContext, and verify them in a real browser. npx @ora-ai/webmcp-verify runs the verification on its own. No signup, no hosted service.

how this was captured

observed via: capture shim (Chromium 148.0.7778.96)
chrome: 148.0.7778.96
capture shim: v1
spec snapshot: 2026-08-26
mode: fast
pages: 1 - entry page only
tools: 20
run time: 20s
finished: Aug 31, 2026

Checked the way in-browser agents discover tools: the top-level document's modelContext registry, read after the page settles.