The 12 pillars — for an MCP tool

MCP (Model Context Protocol) is an open standard, originally published by Anthropic, that lets AIs discover and call your product's actions in a predictable format — instead of scraping your user interface. If you already have a public API (Application Programming Interface), an MCP server is a thin, standardized wrapper on top of it that any MCP-aware AI client can use.

Each pillar has three lines: the plain-English promise, what the platform actually does for your MCP tool today, and the one-click delivery that's coming next. Every abbreviation is expanded the first time it appears.

For the shared background — why 12 pillars, the two jobs (recommendation vs. invocation), and the recommended reading order — see the pillars overview.

• AEOAnswer Engine Optimization

  Make each tool description quotable, so AIs pick the right tool.

  Rewrites the description of each of your tools plus the parameter documentation, in the shape AIs use to select tools: a verb-led one-line summary, a "when to use" line, and a "when not to use" line. Output is a copy-paste patch for your MCP server source code.

  Coming next (one-click delivery)

  Open the tool-description patch as a PR (pull request) directly against your MCP server repository.

• GEOGenerative Engine Optimization

  Stay accurate when AI paraphrases your tool responses.

  Suggests structured, quotable fields to include in every tool response (name, id, canonical_url, last_updated) so that when an AI summarizes your response for the user, your brand and canonical link survive the paraphrase.

  Coming next (one-click delivery)

  Open the response-shape changes as a PR (pull request) in your MCP server repository.

• AAIOAgentic Artificial Intelligence Optimization

  Get NAMED and CHOSEN when a user asks "which tool should I use for X?".

  If your MCP server is the product you sell, AAIO scores how often AIs recommend it in "which MCP server for [job]" prompts. Selection-side signals (recommendation-in-answer) live here; invocation-side signals (the actual tool-calling mechanics) live in the MCP pillar below.

  Coming next (one-click delivery)

  Push authority-page rewrites for your marketing domain into your CMS (Content Management System — WordPress, Webflow, Sanity, Contentful).

• AXOAgent Experience Optimization

  Make each tool response fit the AI's context window.

  Reviews the shape of every tool's response against a typical AI context budget (context = the finite amount of text an AI can hold at once, measured in tokens — small pieces of text). Recommends pagination, trimmed fields, and cleaner error shapes so AIs don't drop your tool from long reasoning loops for being too verbose.

  Coming next (one-click delivery)

  Open the pagination and trim recommendations as a PR (pull request) in your MCP server repository, with unit tests for the new response sizes.

• DAODynamic Agent Optimization

  Adapt tool descriptions and error verbosity per AI client.

  Suggests per-client overrides for tool descriptions, parameter ordering, and error verbosity across the major MCP clients (Claude Desktop from Anthropic, Cursor, Copilot Studio from Microsoft, Continue, Cline, Zed, Gemini Code Assist, Perplexity Tools). Different clients present tools to their model differently, so the same description doesn't land the same way in each.

  Coming next (one-click delivery)

  Ship the per-client overrides as a small config file the platform commits to your MCP server repository as a PR (pull request).

• PULSEPULSE (a pillar name, not an acronym)

  Track your tool's AI visibility every day.

  Asks the AI panel category-open prompts ("use a CRM (Customer Relationship Management system)", "book travel", "query my analytics") to estimate how often your MCP server would be selected. Panel = ChatGPT + Gemini queried directly, plus Claude, Perplexity, Grok, Copilot, Meta AI, and Apple Intelligence simulated on those two using persona techniques — labelled real vs. simulated in the interface.

  Coming next (one-click delivery)

  Direct API (Application Programming Interface) connections to Claude, Perplexity, Grok, and Copilot so those four become real queries instead of simulated ones.

• RIVALRIVAL (a pillar name, not an acronym)

  See which competing MCP tools AIs pick instead of yours.

  Diffs your declared tool catalog against rival MCP servers you name — naming clashes (two tools called "search" fight for the same slot), missing parameters that rivals expose, and scope friction that makes an AI reach for a competitor.

  Coming next (one-click delivery)

  Open the naming-and-parameter fixes as a PR (pull request) in your MCP server repository, and alert on new rival tools that would create a fresh clash.

• PROTOPROTO (short for "Protocol")

  Track updates to the MCP spec and per-host extensions.

  Tracks revisions to the MCP specification itself (features like Streamable HTTP transport, elicitation flows, resource templates, and OAuth authorization profiles) plus per-host extensions from Anthropic, OpenAI, Microsoft, and others. Flags which fields in your current server source will need to change before the next spec revision lands.

  Coming next (one-click delivery)

  Auto-open a PR (pull request) with the spec-compliance fixes ahead of each MCP release.

• BUGBUG (short for "Bug Hunter", a pillar name)

  Catch AIs describing your tools without actually calling them.

  Flags tools an AI describes but never invokes (a common hallucination pattern — the AI convinces the user it has done the thing without actually calling the tool), wrong parameter assumptions, and stale schema patterns that no longer match your live tool.

  Coming next (one-click delivery)

  Auto-open a PR (pull request) with schema and description fixes that reduce the hallucination rate, plus per-AI alerts on new hallucination patterns.

• SIGNALSIGNAL (a pillar name, not an acronym)

  Publish a change feed so AI hosts re-index your tool changes fast.

  Drafts a change-feed payload to publish alongside your MCP server so AI hosts (the applications that host MCP clients — Claude Desktop, ChatGPT, Cursor, and so on) can re-index tool changes without waiting for the next full refresh.

  Coming next (one-click delivery)

  Direct submission to programmatic MCP registries (searchable directories of MCP servers — Smithery, MCP-Index, and vendor-run registries) whenever your change feed updates.

• MCPMCP (Model Context Protocol — an open standard from Anthropic that lets AIs discover and call your product's actions)

  This pillar's home surface. Make sure AIs can find, understand, and reliably call your tools.

  Validates your live MCP endpoint: the shape of the tools/list response (the standard MCP method AIs call to enumerate your tools), the completeness of each tool's JSON Schema (a standard way of describing what parameters a tool accepts), the auth surface (OAuth 2.1 with PKCE — Proof Key for Code Exchange, a security extension that makes OAuth safe for public clients — plus API-key handling), and whether your server works in every major MCP client without per-client patches. Generates ready-to-paste client configs (for example, the small JSON file Claude Desktop uses to register a new MCP server).

  Coming next (one-click delivery)

  Submit your MCP server to the public MCP registries that support programmatic submission with one click; for the rest, generate a copy-paste submission payload.

• SOCIALSOCIAL (a pillar name, not an acronym)

  Consistent developer identity — the trust check an AI host runs before allowing your tool.

  Checks the bio and team page on your developer domain, which AI hosts (Claude Desktop, Cursor, ChatGPT desktop) look at when deciding whether to trust an MCP server enough to expose it to their users.

  Coming next (one-click delivery)

  Open a PR (pull request) against your developer domain's About / team page with the improvements needed to clear the AI-host trust check.