Cybersec — AI Cybersecurity Plugin
180+ pure-Python MCP security tools + 23 pentesting methodology skills for AI coding agents.
Works with: Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI, , , .
接入你的智能体
從發布者安裝。不是 PlugMCP 端點。
遠端 URL 還是本機指令? 兩種安裝方式的差異
來自發布者
180+ pure-Python MCP security tools + 23 pentesting methodology skills for AI coding agents.
Works with: Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot CLI, , , .
Empowers the AI to perform penetration testing, vulnerability assessment, reconnaissance, and security reporting — all through natural language conversation.
Empowers the AI to perform penetration testing, vulnerability assessment, reconnaissance, and security reporting — all through natural language conversation.
Installation differs by harness. If you use more than one, install Cybersec separately for each one.
git clone https://github.com/sukirman1901/cybersec.git
cd cybersec
python3 -m venv mcp/.venv
source mcp/.venv/bin/activate
pip install 'fastmcp>=3.4.2,<4' httpx dnspython 'impacket>=0.12.0'
deactivate
The MCP server needs FastMCP >=3.4.2,<4, httpx and dnspython. Target-specific calls require an initialized and explicitly authorized engagement in the same MCP session. See preflight and audit fixes for setup, result compatibility and enforcement limits.
/plugin install cybersec@claude-plugins-official
/plugin marketplace add sukirman1901/cybersec
/plugin install cybersec@cybersec
Create or edit .mcp.json in your project root:
{
"mcpServers": {
"cybersec": {
"command": "/path/to/cybersec/mcp/.venv/bin/python3",
"args": ["-m", "server"],
"cwd": "/path/to/cybersec/mcp"
}
}
}
/add-plugin cybersec
The plugin auto-registers skills and hooks. You also need to add the MCP server to Cursor's MCP settings.
Available via the official Codex plugin marketplace.
Cybersec in the Security section.+ next to Cybersec and follow the prompts./plugins
cybersec
Install Plugin.gemini extensions install https://github.com/sukirman1901/cybersec
gemini extensions update cybersec
The GEMINI.md file auto-loads the bootstrap context and tool mapping.
copilot plugin marketplace add sukirman1901/cybersec
copilot plugin install cybersec@cybersec
Available in Kimi Code's plugin marketplace.
Open Kimi Code's plugin manager:
/plugins
Go to Marketplace > Cybersec and install it.
Or install directly from this repository:
/plugins install https://github.com/sukirman1901/cybersec
OpenCode uses its own plugin install. The plugin auto-registers the MCP server and all skills — no manual MCP config needed.
Fetch and follow instructions from https://raw.githubusercontent.com/sukirman1901/cybersec/refs/heads/main/.opencode/INSTALL.md
Add to your opencode.json:
{
"plugin": ["cybersec@git+https://github.com/sukirman1901/cybersec.git"]
}
To pin a specific version:
{
"plugin": ["cybersec@git+https://github.com/sukirman1901/cybersec.git#v1.0.0"]
}
Restart OpenCode. The plugin auto-registers the MCP server, skills directory, and bootstrap context.
Install Cybersec as a Pi package from this repository:
pi install git:github.com/sukirman1901/cybersec
For local development:
pi -e /path/to/cybersec
The Pi extension (.pi/extensions/cybersec.ts) loads skills and injects the bootstrap at session startup.
After installation, verify it works by asking your agent:
"scan port 80,443 on example.com" "check SSL for example.com" "enumerate subdomains for example.com" "audit example.com and generate a report"
The AI will automatically load the right skill and use the MCP tools.
On OpenCode, press Tab to switch to the Pentester agent (red), then:
audit example.com
Or invoke subagents directly:
@recon gather intelligence on example.com
@vuln-analyst scan for vulnerabilities
@report-writer generate a report
subdomain_enum, port_scan, http_probe_targetThe AI follows a standard pentesting methodology: Recon → OSINT → Scanning → Vulnerability Analysis → Web Testing → Bug Bounty → AD → Cloud → Password → AI → AI Safety → Desktop → Code Audit → Exploitation → Reporting
For continuous monitoring: CTEM runs in parallel — Discovery → Validation → Prioritization → Remediation in a continuous loop.
User: "Test security for example.com"
The AI will:
cybersec-recon skill → run whois_lookup, dns_lookup, subdomain_enum, crt_searchcybersec-scanning skill → run port_scan, http_probe_target, ssl_check, waf_detectioncybersec-vulns skill → run cve_search, vuln_scancybersec-web skill → run tech_detect, cors_check, dir_bruteforce, auto-CAPTCHA detectionThe Cybersec MCP server provides 180+ security tools via stdio transport using FastMCP.
On platforms that don't auto-register the MCP server (Claude Code, Cursor, Codex, Gemini, etc.), add the MCP server to your platform's MCP configuration:
{
"mcpServers": {
"cybersec": {
"command": "/path/to/cybersec/mcp/.venv/bin/python3",
"args": ["-m", "server"],
"cwd": "/path/to/cybersec/mcp"
}
}
}
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | ./mcp/.venv/bin/python3 -m server
Expected: 180+ tools listed.
MCP tools identify CVE-2026-109618 and CVE-2026-109619 without PHP2SHELL exploit/load probes. The current server registers 209 tools.
php2shell_scan(target="https://your-authorized-host.example") performs normal
GET/HEAD/OPTIONS and optional discovery of five common phpinfo pages. It is not
the broad vuln_scan; no credential checks run. Default 8 requests (hard max 12),
200 KB decoded body limit per response, TLS verified, 30-second whole-scan deadline.
Redirects must keep the same scheme, host and port (max 2 hops); HTTP 429 stops
scanning. Use phpinfo_paths=false to disable diagnostic-page discovery.vuln_scan(target="https://your-authorized-host.example") adds a php2shell
assessment from existing response headers. Existing unrelated scan checks still run.cve_search(service="php", version="8.4.26") includes offline version-based
PHP2SHELL matches even if the NVD lookup fails.php2shell.status is potentially_affected (two probable CVE findings),
not_affected_by_version (advertised version meets its branch fix), or unknown
(insufficient evidence, EOL/future branch, prerelease or conflicting versions).
Fixed baselines: 8.2.35 / 8.3.36 / 8.4.27 / 8.5.12.
Neither a banner nor a Composer PHP constraint proves the deployed runtime.
Check vendor backports and actual serving workers before reporting affectedness
or remediation. A hidden version is not evidence of safety.
The dedicated tool also returns coverage, per-request method/URL/status/error,
presence_signals, and allowlisted phpinfo configuration with source URLs. It
distinguishes access_blocked, php_detected_version_hidden,
redirect_out_of_scope, network_error, rate_limited, budget_exhausted,
timeout, conflicting_evidence, no_php_evidence and version_identified.
version_unclassified means version evidence exists but its branch/build cannot
be classified by the advisory table. These are collection outcomes, separate from
the CVE version assessment. Query strings, fragments and URL credentials are
removed from exported URLs. Requests ask for identity encoding; unexpectedly
compressed bodies are skipped and marked in the request record, not decompressed.
No raw phpinfo pages, cookie values, environment variables or secrets are exported.
session.upload_progress.enabled=Off does not mean either CVE is patched.
The cybersec-vulns and cybersec-verification skills describe these evidence
limits. Restart the MCP server/client after updating. If your client uses an
installed plugin/cache rather than this checkout, update that installation too;
editing this source tree does not update an already-running server or cached skills.
Validation: mcp/.venv/bin/python -m unittest discover -s mcp/tests -p 'test_php2shell*.py' -v.
References: PHP advisory,
VulnCheck,
Ramikan,
JoasASantos.
The reference source was reviewed for banner/diagnostic/presence techniques;
it is not executed or blindly wrapped. Exploit/race/malformed probes, uploads,
secrets-file probing, automatic subdomain expansion and origin bypass are excluded.
Require the corresponding CLI binary. Returns install hint if not found:
Inspired by HackerOne and Pentest-Tools.com — workflow orchestration, findings management, continuous monitoring, and risk scoring.
Automated exploitation, workflow management, OSINT, reporting, and integrations.
The plugin includes 23 methodology skills that guide the AI through structured security testing:
Skills load automatically via intent detection. The AI follows the skill's methodology step by step.
OpenCode agents live in .opencode/agents/. Cursor-compatible mirrors ship in agents/ (declared in .cursor-plugin/plugin.json) with the same roles: pentester, recon, vuln-analyst, report-writer. Cursor also loads rules/cybersec-pipeline.mdc for the default enterprise flow.
Default pipeline: engagement_gate → observe-first / opt-in staging-deep → findings_manager → verification → report_schema_v2 / branded_report → tickets → retest_vuln.
OpenCode-compatible agents are included in .opencode/agents/. These agents have tailored permissions and prompts for different security testing phases.
File: .opencode/agents/pentester.md
Mode: Primary (cycle with Tab: Build → Plan → Pentester)
Color: Red (#DC2626)
Permissions: Full access (edit, bash, read, all MCP tools)
The default pentesting agent. Load this when you want the AI to act as a security tester. It has access to all 180+ MCP tools, can edit files, run bash commands, and dispatch subagents.
Usage:
"audit example.com" or "find vulnerabilities in this codebase"Flow:
User: "audit example.com"
→ Pentester auto-loads cybersec-recon skill
→ Calls @recon subagent for passive intel
→ Calls @vuln-analyst subagent for active scanning
→ Calls @report-writer subagent for report
→ Output: Full penetration testing report
File: .opencode/agents/recon.md
Mode: Subagent (invoke with @recon)
Color: Blue (#3B82F6)
Permissions: Read-only (no edit, no bash)
Passive reconnaissance and OSINT specialist. Gathers intelligence without actively probing the target.
Skills: cybersec-recon, cybersec-osint
Tools: subdomain_enum, dns_lookup, whois_lookup, crt_search, shodan_lookup, wayback_urls, dork_search, ghdb_search, reverse_ip, asn_lookup
Usage:
@recon gather intelligence on example.com
@recon find subdomains for target.com
@recon run OSINT on this domain
File: .opencode/agents/vuln-analyst.md
Mode: Subagent (invoke with @vuln-analyst)
Color: Amber (#F59E0B)
Permissions: Bash + read (no file editing)
Active vulnerability analyst. Scans, fingerprints, and tests for vulnerabilities. Can run bash commands and MCP tools but cannot edit project files.
Skills: cybersec-scanning, cybersec-vulns, cybersec-web, cybersec-bugbounty, cybersec-ad, cybersec-cloud, cybersec-password, cybersec-ai, cybersec-ai-safety, cybersec-desktop, cybersec-code-audit, cybersec-ctf, cybersec-exploit
Tools: port_scan, nmap_scan, ssl_check, vuln_scan, sqli_detect, xss_detect, nuclei_scan, nikto_scan, sqlmap_check, exploit_db_search, cloud_enum, s3_scanner, bloodhound_collect, ldap_enum, bandit_scan, semgrep_scan, prompt_injection, apk_analyze, ipa_analyze, and 80+ more
Usage:
@vuln-analyst scan ports on example.com
@vuln-analyst test for SQL injection in this URL
@vuln-analyst run nuclei scan on target
@vuln-analyst audit this codebase for vulnerabilities
File: .opencode/agents/report-writer.md
Mode: Subagent (invoke with @report-writer)
Color: Green (#10B981)
Permissions: Edit + read (no bash)
Security report writer. Verifies findings and generates professional penetration testing reports with remediation plans.
Skills: cybersec-report, cybersec-verification, cybersec-review
Tools: generate_report, plus all read-only MCP tools for verification
Usage:
@report-writer generate a pentest report from these findings
@report-writer verify these vulnerabilities and write a report
@report-writer create a remediation plan for these issues
Report structure:
# Penetration Testing Report
## Executive Summary
## Scope
## Methodology
## Findings (sorted by severity: Critical → High → Medium → Low → Info)
- Description, Evidence, Reproduction, Impact, Remediation
## Remediation Summary
## References (CVEs, OWASP, etc.)
┌─────────────────────────────────┐
│ PENTESTER (primary) │
│ Tab → Build → Plan → Pentester │
│ Full access, all skills │
└──────────────┬──────────────────┘
│
┌──────────────┼──────────────┐
│ │ │
▼ ▼ ▼
┌──────────┐ ┌─────────────┐ ┌──────────────┐
│ @recon │ │@vuln-analyst│ │@report-writer│
│ read-only│ │ bash + read │ │ edit + read │
│ passive │ │ active │ │ report │
│ OSINT │ │ scanning │ │ verify+write│
└──────────┘ └─────────────┘ └──────────────┘
You can create your own agents by adding markdown files to .opencode/agents/:
---
description: Custom security analyst for your workflow
mode: subagent
permission:
edit: deny
bash: allow
---
Your custom prompt here.
See the OpenCode Agents documentation for full reference.
# Verify MCP server works
./mcp/.venv/bin/python3 -m server
# Should start a stdio MCP server — test with:
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | ./mcp/.venv/bin/python3 -m server
Expected output: 180+ tools listed.
~/.config/opencode/opencode.json for the plugin lineopencode run --print-logs "hello" 2>&1 | grep -i cybersec.mcp.json MCP server config/plugin list should show cybersecsource mcp/.venv/bin/activate
pip install 'fastmcp>=3.4.2,<4' httpx dnspython 'impacket>=0.12.0'
CLI wrapper tools (nikto, sqlmap, etc.) return an error with install instructions when the binary is missing. The core 25+ tools work without any external dependencies.
fastmcp>=3.4.2,<4, httpx, dnspython (installed in .venv)MIT License - see LICENSE file for details.
Empowers the AI to perform penetration testing, vulnerability assessment, reconnaissance, and security reporting — all through natural language conversation.
工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Cybersec 透過 stdio 在本機執行。
上次檢查時間:2026-10-27。 Empowers the AI to perform penetration testing, vulnerability assessment, reconnaissance, and security reporting — all through natural language conversation.