bearstrike-ai - 🐻 BearStrike AI: High-Signal Pentesting Console (M

项目说明

<di="center">
<img src="docs/images/bearstrike-logo-hero.png" alt="BearStrike AI Hero" width="720" />

BearStrike AI

High-Signal Pentesting Console (MCP-First, Model-Agnostic)

![Python]()
![License](LICENSE)
![MCP]()
![Platform](#installation)
![Tools](#tooling)

Build fast, low-noise security workflows with queue-first execution, skills-aware planning, and production-grade runtime state.

</div>

🚀 BearStrike AI Highlights

BearStrike AI is engineered for high-signal pentesting and bug bounty workflows, cutting through noise with intelligent AI orchestration. Here’s what makes it stand out:

🎯 Target Control + WAF Context: Understand your target’s defenses at a glance.

🧠 Smart Planner Panel: AI-dri prioritizes high-value attack paths.

⏱️ Async Jobs + Output Timeline: Monitor scans in real-time with detailed logs.

🛠️ Tool List with Category/Status Filtering: Manage 150+ integrated security tools effortlessly.

🧹 Maintenance Panel for Retention/Cleanup: Keep your workspace clean and efficient.

✨ Featured On:

![MCP Market]()

---

Why BearStrike

Architecture

Core Capabilities

Installation

Quick Start

MCP Client Setup

Tool Use Modes

Common Workflows

Dashboard

Tooling

Skills Playbooks

Data and Retention

Project Structure

Troubleshooting

Legal and Safety

---

Why BearStrike

BearStrike is built for practical bug hunting where signal matters more than scan noise:

Queue-first execution to prevent MCP request stalls.

Request dedupe + response caching to reduce repeated scans and token waste.

Skills-first workflow for planning before offensi.

Durable control plane (SQLite + file artifacts) for stable state across sessions.

Model/client-agnostic MCP compatibility (Claude Desktop, VS Code, Cursor, more).

User-driven hunt mode by default (no forced autonomous chatter loops).

---

Architecture

---

Core Capabilities

1) MCP Control Layer

stdio + sse transports.

Queue-first direct calls with short wait + async fallback.

Concurrency limits, hea cooldown, and tool timeout profiles.

Exact exposure checks via mcp_tool_in.

2) Smart Execution Guardrails

Fingerprint dedupe (tool + target + params + mode + scope_tag).

Cache TTL profiles (low_noise, balanced, aggressive).

Per-target heavy tool controls to reduce rate-limit spikes.

3) Intelligence + Planning

Deterministic endpoint scoring (1-10).

Priority-band dri for high-value paths.

Curated research ingestion and query support.

4) Dashboard Operations

Tool registry with category/status filters.

Async jobs panel with queue state and output preview.

Planner + prioritized endpoints panel.

Cache/dedupe/queue metrics + maintenance controls.

---

Installation

Requirements

Python 3.10+

Linux/WSL recommended for full tool compatibility

Toolchain like nmap, httpx, subfinder, wafw00f, nuclei (as needed)

Setup

Configure

Edit config.json:

---

Quick Start

Full stack (Terminal + Dashboard + MCP)

Start with target

Standalone mode (no MCP/AI loop)

Help

---

MCP Client Setup

BearStrike supports both MCP transports:

stdio for Claude Desktop, Cursor, and VS Code MCP clients.

sse for HTTP/SSE-capable MCP clients.

1) Stdio (recommended)

Run manually:

2) Claude Desktop (Windows + WSL)

Open Claude Desktop MCP config and add:

Then restart Claude Desktop and check De -> Local MCP Servers.

3) Cursor MCP config

Add BearStrike in Cursor's MCP JSON:

4) VS Code MCP config

Use the same mcpServers entry in VS Code MCP settings.

5) Generic local Linux config

6) SSE (HTTP clients)

SSE endpoint:

---

Tool Use Modes

BearStrike supports two dimensions of control:

A) Scan mode (`mode`)

low_noise: safest profile, longest cache TTL, fewer aggressi.

balanced: default profile, practical mix of speed and safety.

aggressive: shortest TTL and faster probing; higher block/rate-limit risk.

Use low_noise for protected targets and bounty programs.

B) Hunt strategy (`strategy`)

adaptive: phased and smart (recommended). Runs 1-2 key tools, analyzes signal, then selects next tools.

di: broader category co while still controlled.

all: queues a wide set of compatible tools quickly (high ).

Default recommendation:

Start: mode=low_noise, strategy=adaptive

Expand: mode=balanced, strategy=di

Only when needed: strategy=all

Inspect mode options (MCP)

Use hunt_options to see current defaults and a strategy/mode .

---

Common Workflows

1) Minimal safe flow (recommended)

1. set_target
2. detect_waf
3. plan_scan
4. full_hunt with strategy=adaptive
5. Track with list_jobs, then job_result
6. Generate report with get_target_report

2) Async-heavy flow (prevents MCP timeout)

Queue: start_tool

Monitor: job_status

Read output: job_result

This is preferred for long tools (subfinder, nuclei, large crawls).

3) Full campaign flow (broad)

Notes:

max_tools=0 means all compatible installed tools.

=false reduces MCP response chatter and token burn.

If rate-limited, switch to strategy=adaptive and mode=low_noise.

---

Dashboard

Open the URL printed at startup (for example ).

Features:

Target control + WAF context.

Smart planner panel.

Async jobs + output timeline.

Tool list with category/status filtering.

Maintenance panel for retention/cleanup.

Main APIs:

GET /api/tools

GET /api/dashboard

GET /api/jobs

POST /api/jobs/start

POST /api/target

GET /api/endpoints/prioritized

GET /api/research/summary

POST /api/research/refresh

GET /api/cache/stats

GET /api/dedupe/stats

POST /api/maintenance/purge

---

Tooling

BearStrike includes 150+ tool profiles across categories:

Recon

Web

Exploit

Cloud

Binary

Forensics

Misc

Core MCP helper tools:

health

list_tools

mcp_tool_in

tools_status

set_target

detect_waf

execute_tool

start_tool

job_status

job_result

list_jobs

plan_scan

smart_scan

full_hunt

score_endpoint

research_refresh

research_query

cache_stats

dedupe_stats

queue_stats

purge_old_data

Broad Campaign Example (`full_hunt`)

Use one call to queue a wide, no-prompt campaign over the main target and discovered subdomains:

Notes:

max_tools=0 means all compatible installed tools.

strategy=all uses pure priority order.

strategy=di spreads co across categories.

---

Skills Playbooks

BearStrike includes a skills-first playbook layer under skills/.
These are designed to be read before hea so MCP clients choose high-signal steps first.

Current skill modules:

planning - pre-scan decision flow and priorities

recon - recon sequencing and endpoint discovery

pentest-tools - tool-selection policy by target and risk

bug-hunting - end-to-end bug hunting methodology

exploitation - PoC quality and false-positi

reporting - structured reporting templates and e standards

research-notes - latest research patterns and bug classes

traffic-proxy - interception workflow (mitmproxy/ZAP-style usage)

xray-suite - xray/crawlergo/headless-browser specific guidance

Recommended MCP behavior:

Load planning + research-notes first.

Use mode=low_noise + strategy=adaptive by default.

Escalate to broader tool sets only after signal is found.

Keep =false in long hunts to reduce response chatter/token burn.

---

Data and Retention

Control plane DB: data/bearstrike.db
Artifacts: reports/output/<target_slug>/...

Retention controls:

Purge runtime data older than N days (recommended: 7 days).

Optional full clear-all reset from dashboard/API.

Examples:

purge_old_data(days=7, include_research=false, =true)

purge_old_data(clear_all=true, =true)

---

Project Structure

---

Identity

<di="center">
<img src="docs/images/bearstrike-emblem.png" alt="BearStrike Emblem" width="360" />
<img src="docs/images/bearstrike-logo-classic.png" alt="BearStrike Classic" width="360" />
</div>

---

Troubleshooting

Only a few MCP tools appear

Start MCP ser compact mode.

Restart your MCP client app (tool schemas can be cached).

Run mcp_tool_in to exposed tools from server side.

MCP timeouts

Use async flow for hea: start_tool -> job_status -> job_result.

Prefer mode="low_noise" for WAF-protected targets.

Avoid parallel hea on same target.

Dashboard port conflict

BearStrike auto-selects next free port and prints final URL.

---

Legal and Safety

Use BearStrike only on systems you own or are explicitly authorized to test.

No unauthorized testing.

Follow program policy (HackerOne, Bugcrowd, TryHackMe, private scopes).

Respect rate limits and safe-disclosure rules.

---

License

MIT License. See LICENSE.

项目信息

开发语言：Python

Star 数：8

技术要点

该项目属于安全软件方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕🐻 BearStrike AI: High-Signal Pentesting Console (MCP-First, Model-Agnostic). Build fast, low-noise security workflows with queue-first execution, skills-aware planning, and 150+ integrated tools.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。

登录后才能参与讨论

下载链接

会员才能下载

登录后开通会员

同分类推荐

暂无同类资源

资源标题	bearstrike-ai - 🐻 BearStrike AI: High-Signal Pentesting Console (M
所属分类	安全软件
发布作者	admin
发布时间	2026-10-05 23:06:14
下载权限	免费下载
附件数量	2 个下载链接
浏览次数	190
讨论数量	0
主题状态	正常讨论中

bearstrike-ai - 🐻 BearStrike AI: High-Signal Pentesting Console (M

项目说明

BearStrike AI

High-Signal Pentesting Console (MCP-First, Model-Agnostic)

🚀 BearStrike AI Highlights

✨ Featured On:

Table of Contents

Why BearStrike

Architecture

Core Capabilities

1) MCP Control Layer

2) Smart Execution Guardrails

3) Intelligence + Planning

4) Dashboard Operations

Installation

Requirements

Setup

Configure

Quick Start

Full stack (Terminal + Dashboard + MCP)

Start with target

Standalone mode (no MCP/AI loop)

Help

MCP Client Setup

1) Stdio (recommended)

2) Claude Desktop (Windows + WSL)

3) Cursor MCP config

4) VS Code MCP config

5) Generic local Linux config

6) SSE (HTTP clients)

Tool Use Modes

A) Scan mode (mode)

B) Hunt strategy (strategy)

Inspect mode options (MCP)

Common Workflows

1) Minimal safe flow (recommended)

2) Async-heavy flow (prevents MCP timeout)

3) Full campaign flow (broad)

Dashboard

Tooling

Broad Campaign Example (full_hunt)

Skills Playbooks

Data and Retention

Project Structure

Identity

Troubleshooting

Only a few MCP tools appear

MCP timeouts

Dashboard port conflict

Legal and Safety

License

项目信息

技术要点

使用与学习建议

免责声明

怎么获取这个资源？

下载链接失效了怎么办？

可以转载或二次发布吗？

下载链接

同分类推荐

A) Scan mode (`mode`)

B) Hunt strategy (`strategy`)

Broad Campaign Example (`full_hunt`)