MCP: add tools for AI-Search / AEO visibility tracking (brand mentions, share of voice)

[manuel-will]

Problem

The MCP server exposes 20+ tools for classic SEO (keywords, backlinks, SERPs, rank tracking, audits), but none of them touch AI-Search / AEO (Answer Engine Optimization). The app already has a full feature area for this — src/server/features/ai-search/services/ (brandLookup.ts, citedSources.ts, promptExplorer.ts, shareOfVoice.ts), backed by src/server/lib/dataforseo/ai.ts — but it's UI-only today.

For an agent-first product this is a gap: an agent can research keywords, audit a site, and track rankings end-to-end, but can't answer "who's winning share of voice in AI answers for this domain, and which pages get cited" without a human opening the AI-Search tab.

Proposal

Add MCP tools that thinly wrap the existing ai-search services (no new DataForSEO integration needed, purely surfacing what already exists):

• get_ai_search_visibility — wraps shareOfVoice.ts + brandLookup.ts: brand mention count and share-of-voice vs. named competitors across a project's tracked prompts.
• get_ai_search_cited_sources — wraps citedSources.ts: which URLs/domains get cited as sources in AI answers for a project's prompts.
• get_ai_search_prompt_results (optional, second priority) — wraps promptExplorer.ts: raw per-prompt AI answers plus citations, for deeper inspection.

Pattern to follow: one file per tool under src/server/mcp/tools/ (Zod input/output schema, handler calling the existing feature service — not the raw DataForSEO client), registered individually in src/server/mcp/server.ts (tools are added one call at a time by design, not via a loop — see the comment at the top of that file). get-domain-overview.ts is a solid template: project-scoped auth via withMcpProjectAuth, market/location resolution, and a credit-cost note in the tool description (DataForSEO's LLM endpoints tend to be pricier than Labs/SERP calls, worth stating explicitly).

Scope: read-only tools only, same risk profile as the existing read tools. No AI-Search config mutation in this proposal.

Prior art

Checked existing issues/PRs before writing this up. No existing issue or PR proposes MCP tools for this area. Two adjacent, currently open items touch the same services and are worth knowing about:

• Add a Cited pages tab to Prompt Explorer #66 / PR feat(prompt-explorer): add a Cited pages tab #85 — "Add a Cited pages tab to Prompt Explorer" (a UI tab on top of citedSources.ts, not an MCP tool). If it merges first, check whether it reshapes citedSources.ts's output before finalizing get_ai_search_cited_sources's schema.
• Site Audit: add AEO/agent-readiness checks (llms.txt, AI crawlers in robots.txt, markdown alternates, linked JSON-LD) #99 / PR feat(audit): AI readiness checks — robots.txt AI crawlers, llms.txt, and Markdown alternates (part 1 of #99) #122 — AEO/agent-readiness checks in Site Audit (llms.txt, robots.txt AI crawlers). Different concern (crawlability vs. answer-visibility tracking), no real overlap.

Notes

Happy to open a PR following the get-domain-overview.ts / list-projects.ts pattern, including tests, once aligned on which of the three tools above are worth building first (get_ai_search_visibility and get_ai_search_cited_sources are the clear priorities; get_ai_search_prompt_results can follow).

[isatimur]

I'll pick this up — starting with the two priority tools (get_ai_search_visibility, get_ai_search_cited_sources), following the get-domain-overview.ts pattern with withMcpProjectAuth, a credit-cost note in the description, and tests mirroring search-console-tools.test.ts. Will note the citedSources.ts finding in the PR: it's a shaping helper consumed by brandLookup.ts's topPages/topQueries output rather than a standalone service, so get_ai_search_cited_sources will call getBrandLookup with the same input schema as visibility and project just the citation-relevant fields. Will hold off on get_ai_search_prompt_results for a follow-up PR per your prioritization.

[isatimur]

Update: PR #148 now includes all three tools — get_ai_search_prompt_results (the deprioritized third tool) is in as of the latest commit, wrapping promptExplorer.ts and defaulting to a single chat_gpt call to keep an unspecified call cheap. All CI green, ready for review.