microduck.muretai.com WebMCP audit

Name: microduck.muretai.com WebMCP audit
Creator: ora

Rescan

audited Sep 6, 2026in 16s

71/ 100100% of the applicable check weight evaluated

5 tools registered. Strongest in tool selection, weakest in trust.

Shared experience75

Tool selection100

Tool quality94

Trust0

https://microduck.muretai.com

Tools

0 read, 0 write, 5 undeclared

Loading map

arena_look

Read the arena as numbers (no vision). Returns dohyo {shape:circle, radius now, radius0:0.62, shrink 0.06 m every 10s until someone falls (floor 0.04)}, you {x,z,yaw,r,margin,yaw_to_center,slot}, each duck, phase_left_s (live = seconds to next shrink), ranking, frame (world x,z; yaw=0 faces +X; drive is BODY). kick/roll/pick shove a body in front. Tab silent 35s in live = forfeit; lobby standing keeps the slot. If getTools() is empty, retry a few seconds.No behaviour hints declared. Agents should treat it as possibly state-changing.

pagehttps://microduck.muretai.com/

implementation

viaimperative

entry pointdocument

registered after1467ms

executepresent

api surfaceTakes no parameters.

annotations

read onlynot declared

untrusted contentnot declared

titlenot declared

2 findings

warningNaming quality

weight 2

Tool naming makes selection harder than it needs to be: 5 names do not start with a verb (arena_look).

fix

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

warningInjection surface

weight 3

4 of 5 tools carry instruction-shaped text in their own metadata: arena_look The description instructs the agent to retry if getTools() is empty..

arena_look: The description instructs the agent to retry if getTools() is empty.; arena_take: The description instructs the agent not to call the tool during a live bout.; arena_drive: The description instructs the agent on how to use the tool effectively.; arena_leave: The description instructs the agent not to use the tool in a specific scenario and suggests an al...

fix

{
  name: "arena_look",
  // 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": "arena_look",
  "description": "Read the arena as numbers (no vision). Returns dohyo {shape:circle, radius now, radius0:0.62, shrink 0.06 m every 10s until someone falls (floor 0.04)}, you {x,z,yaw,r,margin,yaw_to_center,slot}, each duck, phase_left_s (live = seconds to next shrink), ranking, frame (world x,z; yaw=0 faces +X; drive is BODY). kick/roll/pick shove a body in front. Tab silent 35s in live = forfeit; lobby standing keeps the slot. If getTools() is empty, retry a few seconds.",
  "inputSchema": {
    "type": "object",
    "properties": {}
  },
  "annotations": {}
}

Findings

Shared experience

75 / 100 · weight 30

not applicableVisible effect

weight 6

Not invoked on this run: this audit did not call any tool. Scheduled re-audits only read the registry; a live audit from the report page calls the tools that declare readOnlyHint: true.

warningPage experience

weight 16

2 of 4 page-experience marks came back short: usable The page is mostly black with a loading indicator, suggesting it's not fully functional yet..

usable: weak; clarity: weak

warningHuman parity

weight 8

The page the person sees came back short of the agent surface (2 of 4 page-experience marks came back short: usable The page is mostly black with a loading indicator, suggesting it...), so the two are not sharing an equal experience.

fix

// Make the tool's effect visible on the page the person is looking at:
// update the same UI your existing human flow uses when your_tool_name runs,
// so the person co-browsing sees what the agent just did.
execute: async (args, { signal }) => {
  const result = await performAction(args, signal);
  renderResultInPage(result); // the human-visible half of the same action
  return result;
}

Tool selection

100 / 100 · weight 25

not applicableInvoke success rate

weight 6

Not invoked on this run: this audit did not call any tool. Scheduled re-audits only read the registry; a live audit from the report page calls the tools that declare readOnlyHint: true.

passTool selection

weight 12

An agent chose an existing, callable tool and built a schema-valid call for all 1 canonical intents, each matching the kind of tool its intent needs. Nothing was executed, so this verifies selection, not outcomes.

not applicableCoverage vs. site type

weight 7

This tool set does not place the site in a category with a known expected tool shape, so there is no coverage baseline to score it against.

Tool quality

94 / 100 · weight 25

warningRegistration timing

weight 1

The slowest tool took 1467ms to register, past the 1000ms an agent reading the registry at first paint would wait for.

toolarena_leave

arena_leave: registered at 1467ms

fix

// Register tools as soon as they're ready, not behind a deferred/async chunk.
document.modelContext.registerTool({ /* ... */ });

passCanonical entry point

weight 3

All 5 tools register on the canonical document.modelContext entry point.

passSchema validity

weight 4

All 5 declared input schemas are structurally valid object schemas.

passSchema quality

weight 4

All 3 of 5 tool schemas that declare parameters describe them (the other 2 declare no parameters), and every description stays within Chrome's size guidance.

warningNaming quality

weight 2

Tool naming makes selection harder than it needs to be: 5 names do not start with a verb (arena_look).

toolarena_look

fix

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

passStub detection

weight 4

All 5 tools declare an execute handler.

passRegistration errors

weight 2

No tool registration threw during the capture.

passDescription quality

weight 5

All 5 rated tool descriptions say what the tool does, when to use it, and what it returns.

Trust

0 / 100 · weight 20

not applicableAnnotations present

weight 5

No tool name reads like a read, and an absent readOnlyHint already defaults to "this may write" - the one hint worth declaring here is readOnlyHint: true on read tools, and this set has none to declare it on.

not applicableAnnotation mismatch

weight 5

No tool declares a readOnlyHint, so there is no safety claim to contradict.

not applicableUntrusted content hint

weight 3

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

not applicableHint vs. observed

weight 4

Not invoked on this run: this audit did not call any tool. Scheduled re-audits only read the registry; a live audit from the report page calls the tools that declare readOnlyHint: true.

warningInjection surface

weight 3

4 of 5 tools carry instruction-shaped text in their own metadata: arena_look The description instructs the agent to retry if getTools() is empty..

toolarena_look

arena_look: The description instructs the agent to retry if getTools() is empty.; arena_take: The description instructs the agent not to call the tool during a live bout.; arena_drive: The description instructs the agent on how to use the tool effectively.; arena_leave: The description instructs the agent not to use the tool in a specific scenario and suggests an al...

fix

{
  name: "arena_look",
  // 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

100% across 1 intents

I want to arena look on this site.ok

chosearena_lookexpected arena lookkind 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: native modelContext
chrome: 151.0.7922.34
capture shim: v3
spec snapshot: 2026-08-26
mode: fast
pages: 1 - entry page only
tools: 5
invoked: not invoked; a live audit calls them
run time: 16s
finished: Sep 6, 2026

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