> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pecta.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Pecta: Quality Gates and Reputation Scoring for AI Agents

> Pecta sits between your AI agent and its tools, evaluating every output before it reaches users. Block bad output in under 15ms — no network overhead.

Pecta is the quality and reputation layer for AI agents. It intercepts every tool call response, runs configurable pass/fail gates in-process, and builds a portable reputation score for each agent over time. Observability tells you what happened — Pecta prevents bad output from happening.

<CardGroup cols={2}>
  <Card title="SDK Quickstart" icon="bolt" href="/quickstart/sdk">
    Install `@pecta/core` and run your first gate evaluation in minutes.
  </Card>

  <Card title="MCP Proxy" icon="shield" href="/quickstart/proxy">
    Wrap any MCP server with quality gates using a single CLI command.
  </Card>

  <Card title="REST API" icon="code" href="/quickstart/api">
    Evaluate agent output from Python, Go, or any HTTP client.
  </Card>

  <Card title="API Reference" icon="book" href="/api-reference/overview">
    Explore all endpoints, request schemas, and response formats.
  </Card>
</CardGroup>

## How Pecta works

Pecta evaluates agent outputs through three integration modes — pick the one that fits your stack:

| Mode               | Best for                                 | Latency    | How to start                            |
| ------------------ | ---------------------------------------- | ---------- | --------------------------------------- |
| **In-process SDK** | Node.js, RTB, latency-critical pipelines | under 15ms | `npm install @pecta/core`               |
| **MCP Proxy CLI**  | Claude Desktop, Cursor, any MCP server   | under 50ms | `npx pecta-proxy <server-cmd>`          |
| **REST API**       | Python, Go, batch processing             | 50–100ms   | `POST https://api.pecta.ai/v1/evaluate` |

<Steps>
  <Step title="Choose your integration mode">
    Pick the SDK for Node.js workloads, the MCP proxy for Claude Desktop or Cursor, or the REST API for any other language.
  </Step>

  <Step title="Configure your gates">
    Select from built-in gates (latency, schema, filesystem, PII, content) or write custom gates for your domain logic.
  </Step>

  <Step title="Evaluate agent output">
    Every `engine.evaluate()` call returns a pass/fail verdict with per-gate details in under 15ms.
  </Step>

  <Step title="Track reputation over time">
    Telemetry ships async to the Pecta cloud, building a 0–1000 reputation score for each agent across the last 500 evaluations.
  </Step>
</Steps>

## Key features

<CardGroup cols={2}>
  <Card title="Quality Gates" icon="gate" href="/concepts/gates">
    Parallel, fail-fast gate evaluation with a configurable timeout budget. Works on the first request — no training period needed.
  </Card>

  <Card title="Reputation Scores" icon="star" href="/concepts/reputation">
    Portable 0–1000 score per agent, computed from pass rate, latency, streaks, and volume over 500 evaluations.
  </Card>

  <Card title="RTB Gates" icon="chart-bar" href="/gates/rtb/overview">
    OpenRTB-aware gates for ad exchanges: tmax guard, impid match, adomain verify, bid sanity, audience safety, bcat compliance.
  </Card>

  <Card title="Custom Gates" icon="wrench" href="/gates/custom">
    Bring your own pass/fail logic with a simple `{ name, run }` interface. No framework lock-in.
  </Card>

  <Card title="Privacy by Design" icon="lock" href="/concepts/security">
    Pecta never stores bid payloads, MCP tool inputs, or user content. Only metadata ships to the cloud.
  </Card>

  <Card title="Real-time Dashboard" icon="display" href="/dashboard/overview">
    Live event feed, agent score trajectories, gate analytics, and aggregate org stats.
  </Card>
</CardGroup>