# Hacker Bob

> 17 Common Vulnerabilities and Exposures (CVE) IDs across 9 open-source projects. Hacker Bob is an autonomous offensive-security agent for authorized targets. Private beta assessments are operator-supervised and human-reviewed.

## Read first

- [Receipts wall](https://hackerbob.ai/cves)
- [About](https://hackerbob.ai/about.md)
- [Developer guide](https://hackerbob.ai/developers.md)
- [Privacy](https://hackerbob.ai/privacy.md)
- [Contact](https://hackerbob.ai/contact.md)
- [Full agent guide](https://hackerbob.ai/llms-full.txt)
- [Agent instructions](https://hackerbob.ai/agents.md)

## When to use Hacker Bob

Use the public REST API, hosted MCP server, A2A interface, or NLWeb endpoint for cleared public Hacker Bob CVE records and discovery. Use authenticated MCP workspace tools when the user asks to connect an agent, authorize an owned domain, prepare and approve an assessment, launch that approved run, or read results.

Use the local runtime only when the user explicitly asks for a local installation. Keep every action inside the named authorization boundary.

## Safety boundary

Public tools remain read-only. Authenticated workspace tools are bound to one workspace. Domain authorization is required before preparation. Preparation does not launch. Each launch requires a short-lived approval for the exact target and scope; changed or expired approvals are rejected.

## Current capability

The hosted MCP endpoint exposes both public record tools and authenticated product tools. OAuth 2.1 and device authorization are supported. The assessment runtime remains private infrastructure; agents receive state, findings, and released reports rather than runtime credentials.

## Public interfaces

- [OpenAPI 3.1 specification](https://hackerbob.ai/openapi.json)
- [MCP discovery](https://hackerbob.ai/.well-known/mcp)
- [MCP server](https://hackerbob.ai/mcp)
- [OAuth protected-resource metadata](https://hackerbob.ai/.well-known/oauth-protected-resource)
- [Agent onboarding skill](https://hackerbob.ai/.well-known/agent-skills/hacker-bob-onboarding.md)
- [A2A agent card](https://hackerbob.ai/.well-known/agent-card.json)
- [A2A endpoint](https://hackerbob.ai/a2a)
- [API catalog](https://hackerbob.ai/.well-known/api-catalog)
- [Public CVE records API](https://hackerbob.ai/api/v1/cve-records)
- [Agent skills](https://hackerbob.ai/.well-known/agent-skills/index.json)
- [API-scoped agent guide](https://hackerbob.ai/api/llms.txt)
- [Documentation-scoped agent guide](https://hackerbob.ai/docs/llms.txt)
- [Developer-scoped agent guide](https://hackerbob.ai/developers/llms.txt)
- [Agent Plugin manifest](https://hackerbob.ai/plugin.json)
- [Schema Map](https://hackerbob.ai/schemamap.xml)
- [Public records JavaScript SDK](https://hackerbob.ai/sdk/hacker-bob.mjs)

## Install the assessment runtime

```sh
npx -y hacker-bob@latest install .
```

- [Source](https://github.com/vmihalis/hacker-bob)
- [npm](https://www.npmjs.com/package/hacker-bob)
- [Apache-2.0 license](https://www.apache.org/licenses/LICENSE-2.0)

## Authorization

Confirm authorization before any assessment. Source visibility is not permission to probe a deployed system. Do not widen the registrable domain or repository scope named by the operator.

Last revised 2026-09-05.