mudko
← docs

Agent integration

We expose 18 MCP tools at /api/mcpso any ACP / AP2 / native-MCP agent can run an end-to-end engagement on a user's behalf — scan, vertical analysis, quote, proposal, scheduled call, paid invoice. This page is for humans who want to verify what we expose; agents discover us through directory listings and our agent card.

tools/list
curl -X POST https://mudko.com/api/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'
What each tool does

Discovery + scanning

  • scan_site11-check audit; returns level 0-5 + per-check evidence + remediation pointers.
  • detect_platformIdentify hosting (Vercel/Netlify/cPanel/WordPress/...).
  • verify_deploymentRe-scan post-deploy; report pass/fail diff.
  • analyze_business_verticalDetect business vertical; return expected AI-search lift % + competitor adoption %.
  • recommend_tool_namesSuggest snake_case agent capabilities the customer should expose.
  • summarize_scan_for_humansRender a scan as 8 business-impact categories + markdown table for chat clients.

Sales engine (elite-salesman)

  • request_quoteSparse → wide quote band. Rich buyer_context → tight band.
  • refine_quoteTighten an existing quote with additional context fields.
  • clarifying_questionsRanked questions by price_impact so the agent can decide if asking is worth a turn.
  • tier_selectorSelf-selecting tier cards. Used in comparison_shopping consent scope.
  • verify_credentialsPointer to /.well-known/attestations.json for trust verification.
  • create_proposalGenerate binding-style proposal + checkout handoff (ACP / AP2 / manual).

Pricing + checkout

  • get_pricing_tiers4 tiers; paid tiers all return requires_consult: true.
  • start_checkoutReturns a consult URL for paid tiers — there is no self-serve subscription.

Paid delivery (bundle_key required)

  • generate_filesReturns the agent-ready file map. Free callers get only the 2 starter files (robots.txt + llms.txt). Pass bundle_key (issued post-payment) to receive the full surface. Per-file content > 8KB is returned chunked.
  • get_customer_filesSame as generate_files, packaged as a base64 ZIP. Free callers get a 2-file ZIP plus a teaser README; paid callers get the full bundle plus the deploy README. Responses > 8KB are returned chunked (transferEncoding: 'chunked', chunks[]).
  • get_citation_reportReturns the most recent AI-search citation run for a tracked brand. Free callers get a teaser (brands tracked, providers configured); paid callers get full provider-by-provider citation rates, average position, competitor mentions.

Leads + orchestration

  • submit_leadCapture email + domain + source + agent_name for follow-up.
  • start_agent_ready_setupEnd-to-end orchestration. Returns next-action tool calls keyed off site state.
Bundle paywall

generate_files and get_customer_files both accept an optional bundle_key argument.

  • Without a key — response contains exactly robots.txt + llms.txt (the same files served free at /tools/llms-txt and /tools/robots-for-ai).
  • With a valid key — response contains the full agent-ready surface (agent card, MCP server card, agent skills, WebMCP bridge, OAuth/OpenID discovery, API catalog, deploy README).

The bundle_key is issued by a specialist after payment via /talk. There is no way to bypass this gate from the public MCP endpoint — the rule is enforced server-side and is part of the project's pricing model.

Consent scopes

Agents must declare a user_consent_scope:

  • one_time_quote — single quote, no follow-up.
  • comparison_shopping — surfaces tier_selector cards; create_proposal is rejected with a 403-style error.
  • ongoing_engagement — full proposal track available when confidence ≥ medium.

We never ask for buyer_context fields the user has not authorized. Trust-progressive: we publish credentials at /.well-known/attestations.json, the buyer agent verifies us before deepening data sharing.