Skip to content

Win the agent decision.

AI agents now choose which software does the job. InvokeRank shows which products they actually choose at every step from directory listing to finished task, with the evidence behind every number.

  • Free
  • No email to see the result
  • An MCP server, plugin, website or product name

Prefer a guided baseline?

Agent directories, registries and tool descriptions flow into InvokeRank, which follows each request through five stages: surfaced, selected, invoked, executed and completed.{ }Claude directoryChatGPT appsTool descriptionsMCP registryCompetitorsSURFACEDSELECTEDINVOKEDEXECUTEDCOMPLETED
01 What we watch

Ten directories and registries. 185,553 unique MCP servers. One index.

Each server counts once, however many directories list it, and every count is live from InvokeIndex with its source attached.

Collectors last ran Glama, PulseMCP and mcp.so listings arrive through MCP Toplist’s open dataset, used under CC BY 4.0.

110,189servers listed
Via MCP Toplist’s dataset
44,386remote MCP servers

Servers with a public endpoint an agent can call.

61,999servers listed
43,753brands resolved

Matched across directories by verified domain evidence, never by name alone.

22,368servers listed
Via MCP Toplist’s dataset
7,290brands on two or more hosts

Most products are listed on one host only.

18,209tool lists captured

Servers whose tools an anonymous probe listed, with every change kept.

20,224servers listed
Via MCP Toplist’s dataset
Google DriveMicrosoft LearnContext7AmplitudeMermaid ChartSigmaAWS MarketplaceQuartr
NotionDocusignClinical TrialsCalendlyLushaAccuWeather®tldvZillow
Microsoft 365Windsor.aiExpediaKiwi.comAWS MCPSlidesGPTSlickTripWolfram

The tools agents can call

185,553unique MCP servers

Across ten major directories and registries, each with its sources and history.

Open the index
SupabaseApollo.ioNetlifyWixPDF ViewerAutodesk Product HelpArticulateDeel
ShopifyConsensusSuperhuman MailAdobe Marketing AgentTrivagoGoDaddyMercuryFiverr
GranolaSupermetrics Marketing AnalyticsExaRampBigdata.comEgnyteParallel SearchActiveCampaign
02 How we measure

Measured, not guessed

Agent distribution is full of accepted wisdom that doesn’t hold up. Here’s what the method is built on.

Listing is not distribution

A directory listing puts you on the shelf. It says nothing about whether an agent reaches for you when a customer asks for the job your product does.

A description is not a decision

Agents read your tool names, descriptions and schemas, then weigh them against every competitor that could do the same job, and against answering on their own. None of that shows up in your analytics.

One run is not a measurement

Ask an agent the same question twice and it can pick a different tool. We repeat every request, report a probability with an interval, and report overlapping results as tied.

One host is not the market

ChatGPT, Claude, Cursor and Codex each find and rank tools their own way. Winning on one tells you little about the others, so each host is measured on its own terms.

Gaming the model does not work

Copy written to trick a model into choosing you gets caught by directory review and by the next model update. We only recommend changes that make your tool genuinely easier to choose correctly, and rankings are never for sale.

Read the methodology
app.invokerank.com/acmeSample data
InvokeRank62 (55 to 69)Completed tasks per 100 eligible requestsVerifiedR + S + sandbox E/C
Selection share29%Claude, 30 days
Where you loseSelectedBiggest drop in the funnel
Overview: how often agents finish a task through you
03 Competition

Do you know who agents choose instead of you?

There’s a difference between thinking agents choose you and knowing. InvokeRank runs the requests your category depends on through the OpenAI and Anthropic APIs, labelled as API runs, and shows where you win, where you lose, and to whom.

The window shows sample data. Real category benchmarks replace it only once they pass the eligibility and evidence gates in the methodology.

04 Action

From “are we chosen?” to “what do we change?”

05 Monitoring

Know when the decision moves

Models update and competitors ship new tools. InvokeSignal reruns your requests on a schedule and alerts you only when a change falls outside its interval, with the evidence attached.

Sample data
Competitor A added 3 tools29 Sep, two days before your share fell
Selection fell beyond noiseSync contacts from email, 38% to 29%
Likely cause foundTheir description names email matching
To email or Slack, with the evidence attached
Why it matters One request, one winner

Agent search is hard to win.

Ask an agent to get a job done and it weighs every tool that could do it, and whether to answer on its own, then picks one. There is no results page and no second click. InvokeRank shows who wins that decision, and why, then measures whether each change you make moves it.

  • Six stages from shortlist to outcome
  • A 90% interval on every number
  • OpenAI and Anthropic APIs, labelled as API runs
Illustration: the same request goes to an agent three times. It weighs your product, two competitors and answering on its own, and picks Competitor A, then no tool, then your product.

Illustration, not data

Free tool Custom GPT retirement

Custom GPTs retire on 11 December. Their Actions do not move.

Instructions and knowledge become a plugin. Custom Actions have to be rebuilt as an MCP server. Paste your GPT and get the plugin package, with a working server, in a minute.

25Days until retirement
0Custom Actions that transfer