MCP server

The AI brand-tracking MCP server, backed by the same REST API. Talk to your agent normally. Zero integration code.

Most "AI visibility" tools assume you sit in their dashboard. Cite42 doesn't. Install the MCP server in Claude Code, Claude.ai, Cursor, or Codex, your agent gets 3 tools: cite42_search, cite42_rankings, and cite42_citations. Ask in plain English. The agent picks the right tool.

$1 free on signup
Credits never expire
No subscription
MCP native

Get $1 free, try it Install MCP

POST /v1/searchlive

1. You ask your AI, in plain English
How is Penflow doing in AI search this week?

2. The agent picks the right cite42_* tool itself
→ calls cite42_rankings across all 4 models

3. You get back, in plain English
Mentioned 71% of the time — strongest on Perplexity.

WHAT YOU CAN ASK

Ask in plain language. Get structured JSON.

Each is one call against the endpoint, ask in plain English through the MCP, or hit the API directly. Same key, same credit balance.

You ask“How is Penflow doing in AI search?”

You get{ "tool": "cite42_rankings", "mention_rate": 0.71, "cost": "$0.28" }

You ask“Which URLs of mine does AI cite?”

You get{ "tool": "cite42_citations", "cited": 3, "cost": "$0.20" }

You ask“What does every model say about us?”

You get{ "tool": "cite42_search", "models": 4, "cost": "$0.32" }

Why MCP-native beats custom glue

Your agent should call AI-visibility tools, not wait for you to wire them.

The hard part of giving an agent brand-tracking powers was never the data, it was the plumbing. You write a tool wrapper, hand-roll the OpenAI, Anthropic, Google, and Perplexity calls behind it, enable web search per provider, reconcile four citation shapes, and keep all of it from rotting every time a provider ships a breaking change. Cite42 ships that as an MCP server instead.

Install @cite42/mcp once and your agent gets three tools, cite42_search, cite42_rankings, and cite42_citations, each fanning out across ChatGPT, Claude, Perplexity, and Gemini. You describe the outcome ("how is Penflow doing in AI search this week?") and the agent picks and chains the right tool. Same key, same per-call pricing as REST, no SDK to babysit.

Three tools the agent owns: cite42_search, cite42_rankings, cite42_citations.
One install line works in Claude Code, Claude.ai, ChatGPT, Cursor, and Codex.
Identical prompts replay from a cross-customer cache at $0.

AI MODELS (4)

Every model, one API key.

One call reaches all 4 AI models, ChatGPT, Claude, Perplexity, and Gemini, or pass “models” to narrow to the ones you need. Billed per model queried.

ChatGPT
OpenAI
Web search on
The highest-traffic AI surface
Claude
Anthropic
Web search tool
Cited, source-grounded answers
Perplexity
Perplexity Sonar
Web-grounded
Always answers from live sources
Gemini
Google
Google Search grounding
Google's own model layer
Google AI Overviews
Google Search
Live SERP
The AI summary atop Google results

PRICED FOR REAL WORK

Real scenarios. Real numbers.

Real workflows, real call counts, real totals, see what this looks like before you commit. Credits never expire.

agent builder
Ship a GEO feature without an SDK
Wire the MCP server into your product agent so users ask "how am I doing in AI search" and it runs cite42_rankings + cite42_citations. A typical session: 6 calls · ~$1.56, billed per call.
Claude Code power user
Run an ad-hoc AI-visibility audit
From your editor, tell Claude to rank your brand on 20 prompts and pull the cited URLs. The agent fans out cite42_rankings across all four models, then cite42_citations on the winners. 25 calls · $5.40 total.
automated workflow
Nightly agent watch on a schedule
A scheduled agent calls cite42_rankings on the same 20 prompts every night and pings Slack on a drop. 20 calls/night across all four models · ~$5.60/day, no agent seats, no MCP markup.

vs. wiring AI calls into your agent by hand

An MCP server the agent uses, or custom tool glue you maintain forever.

You can hand-build agent tools that call each AI provider, parse citations, and meter spend yourself. It works until web search config, four response shapes, and provider breaking changes become a permanent maintenance tax on your agent. Same outcome, both ways.

	Cite42	Hand-built agent tools
Integration	One install line, three tools the agent picks itself	Write and register a tool wrapper per provider, then maintain it
Model coverage	Every tool fans out to ChatGPT, Claude, Perplexity, Gemini	Glue four SDKs together and merge the results yourself
Citations + web search	Web search on per model, citations normalized to { url, title }[]	Enable search per provider, reconcile four citation shapes
Billing	One prepaid key, metered per model per call, no agent seats	Track four invoices and rate limits across providers
Repeat queries	Cross-customer cache replays identical prompts at $0	Pay full provider price every run
Cost, 4-model rankings call	$0.28 ($0.07 per model), or $0 on a cache hit	Four providers summed + the eng time to keep tools alive

PRICING

Pay per call. $25 minimum deposit. Credits never expire.

This page's endpoints below. See /pricing for the full per-endpoint table.

Endpoint	Price	What you get
POST/v1/search	$0.08/model	Raw prompt across all models, called via the MCP tool cite42_search
POST/v1/rankings	$0.07/model	Brand rankings, called via the MCP tool cite42_rankings
POST/v1/citations	$0.05/model	Citation lookup, called via the MCP tool cite42_citations

See all endpoints and the model coverage table →

KEEP EXPLORING

Related endpoints

Same API key, same credit balance, different question.

AI searchAI searchOne prompt in, every model's answer out.Explore ai search AI search visibilityAI brand rankingsMention rate, position, and share of voice.Explore ai brand rankings AI citation trackerAI citationsCitations are the new backlinks.Explore ai citations

QUESTIONS

Questions about agent + mcp.

Anything that implements the MCP spec, Claude Code, Claude.ai (custom connectors), Cursor, Codex, Windsurf, Zed. Setup is identical: paste the install command, restart, tools appear.

No. The same cite42_live_... key works for both REST calls and MCP. The MCP server reads it from the CITE42_API_KEY env var you pass in the install command.

Exactly the same as REST. Each MCP tool wraps a REST endpoint. Calling cite42_rankings is identical pricing to POST /v1/rankings. No agent-seat tax.

Yes. Give the agent its own API key backed by a prepaid balance and it can call any endpoint, or any MCP tool, on its own. Each call is metered and debited from that balance, so you set the budget and the agent runs against it unattended.

Three: cite42_search (raw prompt passthrough across all models), cite42_rankings (brand mention rate and position), and cite42_citations (which models cite a URL). One per live REST endpoint.

READY WHEN YOU ARE

Add Cite42 to your agent.

One line install. $1 free credit to test before you deposit.