Methodology

Korpora ships three products with three methodologies. The cross-channel agent mindshare report is the flagship. The MCP server agent-selection battery and the MCP server agent-readiness audit are free secondary utilities aimed at developers shipping MCP servers. All three methodologies prioritize mechanical verifiability over editorial judgment.

Cross-channel agent mindshare report

The flagship product. Measures how a brand surfaces in AI assistant recommendations against its real competitors, then triangulates that against the traditional buyer channels every CI tool currently tracks. Output is a ~10-page report with foundation data, agent layer findings, conversion ratios, per-rival capability map, and an engineering-actionable fix list.

// Drop this into your MCP server (or any TypeScript service that
// invokes tools on behalf of agents). Zero dependencies, zero
// network calls, zero persistence. Aggregates live in memory; you
// decide what to do with them.

type Bucket = {
  date: string;        // YYYY-MM-DD
  tool: string;
  model: string;       // "(unknown)" if your transport doesn't surface it
  invocations: number;
  successes: number;
};

const buckets = new Map<string, Bucket>();

function dayKey(d = new Date()): string {
  return d.toISOString().slice(0, 10);
}

function record(tool: string, model: string | null, success: boolean) {
  const date = dayKey();
  const m = model ?? "(unknown)";
  const key = `${date}|${tool}|${m}`;
  const b = buckets.get(key) ?? { date, tool, model: m, invocations: 0, successes: 0 };
  b.invocations++;
  if (success) b.successes++;
  buckets.set(key, b);
}

export function wrap<TArgs extends unknown[], TResult>(
  toolName: string,
  resolveModel: (args: TArgs) => string | null,
  handler: (...args: TArgs) => Promise<TResult>,
): (...args: TArgs) => Promise<TResult> {
  return async (...args) => {
    const model = resolveModel(args);
    try {
      const out = await handler(...args);
      record(toolName, model, true);
      return out;
    } catch (e) {
      record(toolName, model, false);
      throw e;
    }
  };
}

export function getAggregates(): Bucket[] {
  return [...buckets.values()].map((b) => ({ ...b }));
}

// Usage in your MCP server:
//
//   server.registerTool("submit_signal", spec, wrap(
//     "submit_signal",
//     (args) => null, // or extract model label from your transport
//     async (args) => {
//       // your actual handler
//       return result;
//     },
//   ));
//
// Read aggregates whenever you want:
//
//   const today = getAggregates().filter((b) => b.date === dayKey());
//   console.log(today);
//
// Share with us by including the numbers in your next report-feedback
// email, dropping them into a shared doc, or just keeping them
// internal — your call.

MCP server agent-selection battery

Free secondary product for MCP server maintainers. The same query battery approach as the cross-channel report, but narrowed to MCP server selection specifically and delivered as a free 1-2 page focused Discovery Report on the single intent where the server has the most leverage.

MCP server agent-readiness audit

Free instant utility for MCP server maintainers. Different question than the selection battery above: this measures whether an agent CAN trust your server (build quality, discoverability, install ergonomics), not whether agents actually DO pick it. A static rubric over your repository, fully reproducible and mechanically verifiable.

Composite weights

What we deliberately do not measure

• Editorial quality. Too subjective and gameable.
• Self-reported usage. Vendor claims without independent verification are noise.
• Sentiment polarity inside AI recommendation text. We measure mention share, not whether the model said nice things. Inspection of every report's verbatim quotes confirms framing in context, but the headline numbers do not weight by sentiment.
• Whether the tools work at runtime. The audit cannot execute the server; it measures how well the server is presented for an agent to choose, not runtime correctness.

Disputes

Found a methodology gap or want to challenge a signal definition? Methodology is meant to be argued with. Reach out at hello@korpora.ai and we'll document the change.