Agents
Signals are simulated Analyze
Agents · WebMCP

Give your coding agent permission to say no.

BuildCheck is the pre-build decision layer: should we build, what must we validate, and what is the smallest thing worth coding?

Checking WebMCP…
Experimental open standard

Connect BuildCheck to the agent viewing this page.

The native document.modelContext API is feature-detected. In a supported secure browser, the seven tools below are registered against this live page and current demo session. No remote MCP server is being simulated.

Step 1User intent
Step 2Coding agent
Step 3BuildCheck evidence
Step 4Build / Validate / Pivot / Kill
Step 5Minimum code worth building
Challenge-ready setup

Use the supported Site tools environment.

Use the latest ChatGPT desktop app with GPT-5.6 Sol or Terra. Keep Site tools enabled in Settings → Browser → Permissions.

7/7 tool contracts tested
  • Open this /agents page in the built-in browser
  • Inspect Available site tools in the address bar
  • Run the demo prompt before allowing any code
  • Availability still depends on the current OpenAI rollout and workspace type

Available site tools

Narrow inputs, structured JSON outputs and explicit read/write annotations

Experimental WebMCP
validate_ideawrite

Analyze an idea, persist it in the current workspace, and use an external source only with explicit allow_external_lookup consent.

“Should I build an AI CRM for freelancers?”
roast_idearead

Challenge an existing project or an idea using available evidence and clearly labeled assumptions.

“Roast project generic-ai-crm.”
get_project_analysisread

Return the latest complete analysis for a project owned by the current session.

“Get my BuildCheck analysis for InvoiceFlow.”
generate_validation_mvpread

Reduce a project or idea to the cheapest testable scope and machine-readable success criteria.

“What is the cheapest way to validate it?”
estimate_build_costread

Return directional hour and AI-token ranges plus scope risk factors.

“Estimate the full build cost for RefundOps.”
find_opportunitiesread

Filter opportunities available in BuildCheck; demo mode uses the curated seed feed.

“Find B2B ecommerce opportunities above 80.”
evaluate_before_buildread

Advise whether a full build is justified and return a structured warning plus smaller alternative when it is not.

“Should I build the full product now?”
Try the challenge story

One prompt, before any code.

Open this deployed page in ChatGPT’s in-app browser, inspect Site tools in the address bar, then ask:

Should I build an AI CRM for freelancers? Use BuildCheck before writing code.
Current boundary

Page-bound, not background MCP.

  • Requires a secure context and supported browser
  • Tools belong to the open page
  • Demo state stays in this browser
  • Supabase continuity is schema-ready, not connected
Official WebMCP docs

Agent activity

Local structured audit stream — no prompts, secrets or raw inputs are logged

No agent calls yet

Activity appears here after a compatible browser invokes a BuildCheck site tool.