kali-mcp
kali-mcp is a comprehensive AI-driven penetration testing platform that exposes over 120 MCP tools wrapping 27+ Kali Linux security tools, enabling autonomous or guided offensive security engagements from reconnaissance through exploitation and reporting. Its capabilities include:
Network Scanning & Discovery: Nmap scans (host discovery, port/service/OS detection, vulnerability, aggressive), fast masscan+nmap combinations, netcat port checks, and banner grabbing.
Reconnaissance & OSINT: Subdomain enumeration (subfinder, Amass, gobuster DNS), OSINT gathering (theHarvester), WHOIS/DNS lookups (dig, zone transfers), and continuous recon sweeps with asset change detection.
Web Application Testing: Vulnerability scanning (Nikto, Nuclei, WPScan), directory/vhost/file brute-forcing (gobuster, ffuf), crawling, HTTP request crafting, form submission, link/text extraction, and visual screenshots (gowitness).
Exploitation & Post-Exploitation: SQL injection (sqlmap), password brute-forcing (Hydra), Metasploit integration (search, run, msfvenom payload generation), exploit searching (searchsploit, CVE-to-exploit mapping), automated scan-to-exploit chains, SSH command execution, privilege escalation enumeration, and SMB/Windows enumeration (enum4linux, smbclient).
Traffic Analysis: PCAP credential extraction, protocol hierarchy analysis, and tshark queries.
AI-Optimized Workflows: Parallel multi-tool scans (scan_host, scan_web), intelligent finding triage with attack path analysis, and correlation of findings into compound attack chains.
Engagement & Asset Management: Full engagement lifecycle, scoping with allow/deny targets, persistent asset inventory, and scan-to-scan asset diffing.
Credential Management: Encrypted vault for storing, listing, reusing, and deleting discovered credentials.
Reporting & Validation: Automated report generation (Markdown/HTML) with attack chains and remediation advice, structured finding extraction, and a validation workflow to confirm or dismiss findings.
Job & System Management: Asynchronous job handling (submit, status, output, cancel), server health checks, file artifact management, and runtime scope enforcement.
This platform fully automates the penetration testing lifecycle, transforming any MCP-capable AI into a powerful offensive security assistant.
Provides tools for running Kali Linux security tools, enabling AI agents to perform penetration testing, vulnerability scanning, and attack chain analysis on authorized targets.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@kali-mcp Scan 10.10.10.5 and tell me what's worth attacking."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
๐ kali-mcp
Your AI, holding a full Kali Linux toolkit.
kali-mcp turns any MCP-compatible AI assistant into a hands-on penetration testing partner โ it drives 27+ real Kali security tools through 123 AI-callable actions, and hands the AI clean, verified, structured findings instead of raw terminal noise.
123 MCP tools ยท 27+ Kali binaries ยท autonomous hunt mode ยท verification oracles + interactsh OOB + IDOR/access-control + race & secrets scanners ยท 7 attack-chain templates ยท ~450 tests
Quick Start ยท Documentation ยท Tool Reference
You: "Scan 10.10.10.5 and tell me what's worth attacking."
AI: โ scan_host("10.10.10.5")
โ open ports 22, 80, 445 โ fans out nikto + gobuster + nuclei + enum4linux in parallel
โ 14 findings extracted, verified, and deduplicated
โ attack chain found: Exposed .git โ leaked creds โ Admin Panel
โ next moves suggested: hydra on SSH, sqlmap on the login formYou describe intent. The AI drives the tools. kali-mcp makes the results trustworthy.
โจ Why kali-mcp
๐ง Structured, not raw. Every scan result becomes a clean finding (
host,severity,confidence,evidence) โ never a wall of terminal text. โ The Finding Pipeline๐ฏ Low false positives. Findings are actively re-verified (soft-404 baselines, catch-all clustering,
.git/.envcontent proof), evidence-anchored, and cross-tool corroborated. The AI can't inflate what the tools didn't prove. โ False-Positive Reduction๐ Impact, not just bugs. Individual findings are correlated into named attack chains with ready-to-paste narratives. โ Attack Chains
๐ฐ๏ธ Change detection. Passive recon sweeps diff against previous runs and surface newly exposed assets โ where the bounties are. โ Continuous Recon
๐๏ธ Memory. A persistent asset inventory and full engagement lifecycle, from scope to client-ready report. โ Engagements
๐ Safety-first. stdio-only (no exposed network), scope allow/deny enforcement, argument-injection guards, an encrypted credential vault, and a full audit log. โ Security Model
Related MCP server: MCP Kali Server
๐ Quick Start
# 1. Clone & install
git clone https://github.com/Neeraj829784/kali-mcp.git
cd kali-mcp
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
# 2. Verify
python3 -c "from server import mcp; print('Ready')"Then point your AI client at it:
{
"mcpServers": {
"kali-mcp": {
"command": "/path/to/kali-mcp/.venv/bin/python",
"args": ["/path/to/kali-mcp/server.py"]
}
}
}Ask it to run server_health() to confirm the tools are installed, then
scan_host("<your-authorized-target>").
Full setup โ including the security tools kali-mcp drives โ is in the Installation guide.
๐ Documentation
Everything lives in docs/:
Start here | Concepts | Scope & assets | Reference |
โ๏ธ Responsible use
kali-mcp runs real offensive tooling. Only test systems you own or are explicitly authorized to test. Use programs to encode your authorization boundary and stay inside it. You are responsible for how you use this tool.
๐ License
MIT โ see LICENSE.
Available Tools
101 toolsamass_enumA
In-depth subdomain enumeration using OWASP Amass. domain: target domain (e.g. 'example.com') passive: passive-only mode (no active probing) โ safer, faster brute_force: enable brute-force subdomain discovery (slow) timeout_mins: max runtime in minutes (default 5)
Inputs: domain (required), passive, brute_force, timeout_mins.
analyze_attack_chainsA
Correlate current findings into multi-stage attack chains.
Shows how individual low/medium findings combine into high-impact compound attack paths (e.g. SQL injection + exposed SSH = credential theft + system access). Call this at any point during an engagement to understand compound risk.
host: focus on a specific host (empty = all hosts) min_severity: minimum finding severity to consider โ info, low, medium, high, critical min_confidence: minimum finding confidence to consider โ low, medium, high Returns: list of chains with name, escalated severity, narrative, steps, affected hosts.
Inputs: host, min_severity, min_confidence.
analyze_findingsA
Analyze and prioritize all findings from recent scans. Returns a structured triage report designed for LLM decision-making. Identifies: attack paths, quick wins, critical issues, credential reuse opportunities.
host: focus on specific host (empty = all hosts) min_severity: minimum severity to include โ low, medium, high, critical min_confidence: minimum confidence to include โ low, medium, high max_items: max findings to analyze (default 20)
Returns: prioritized attack paths, quick wins, and recommended next actions.
Inputs: host, max_items, min_severity, min_confidence.
asset_diff_runsA
Compare two scan runs and report what changed between them. Reports new / disappeared / changed hosts, services, and vulnerabilities. prev_run_id: the earlier (baseline) run id curr_run_id: the later run id
Inputs: curr_run_id (required), prev_run_id (required).
asset_get_hostB
Get full details for a specific host.
Inputs: host_ip (required).
asset_latest_changesA
Show what changed between the two most recent completed runs. This is the core continuous-recon view: 'what's new since last time?' program: focus on one program (empty = all runs across programs) Returns: change report + a 'new_assets' list flagged as high-priority (newly deployed assets are often less tested).
Inputs: program.
asset_list_hostsB
List all known hosts. status: filter by status (new, scanned, confirmed, remediated, false_positive) min_scan_count: minimum number of scans limit: max results
Inputs: limit, status, min_scan_count.
asset_list_runsA
List recorded scan runs (recon cycles), newest first. program: filter by program name (empty = all) limit: max results
Inputs: limit, program.
asset_list_servicesB
List services, optionally filtered by host or service name.
Inputs: host_ip, max_port, min_port, service_name.
asset_list_vulnerabilitiesB
List vulnerabilities, optionally filtered by host or severity. min_severity: info, low, medium, high, critical status: unconfirmed, confirmed, false_positive, remediated
Inputs: limit, status, host_ip, min_severity.
asset_mark_hostB
Mark a host with a status (new, scanned, confirmed, remediated, false_positive).
Inputs: status (required), host_ip (required).
asset_mark_vulnC
Mark a vulnerability with a status.
Inputs: status (required), vuln_id (required).
asset_searchA
Search assets by IP, hostname, service, or CVE.
Inputs: query (required).
cancel_jobADestructive
Cancel a running job and kill its process group.
Inputs: job_id (required).
check_binaryA
Check if a specific binary is installed and where it is. name: binary name e.g. 'nmap', 'sshpass', 'msfconsole'
Inputs: name (required).
creds_deleteB
Delete a credential from the vault by its ID.
Inputs: cred_id (required).
creds_listA
List stored credentials, optionally filtered by host or service. host: filter by host IP/hostname (empty = all hosts) service: filter by service type e.g. 'ssh', 'http' (empty = all) Returns credentials with decrypted passwords/hashes for use in further attacks.
Inputs: host, service.
creds_storeA
Store discovered credentials in the vault (passwords/hashes encrypted at rest). host: target host IP or hostname username: discovered username password: plaintext password (leave empty if only hash available) hash: password hash (e.g. NTLM, bcrypt) service: service type e.g. 'ssh', 'http', 'smb', 'ftp', 'mysql' port: service port source_tool: which tool found this (e.g. 'hydra', 'sqlmap', 'manual') notes: any additional context
Inputs: host (required), username (required), hash, port, +4 more.
creds_useA
Get the most recently discovered credential for a host/service. Use this before attacking a service to check if we already have valid creds. host: target host service: optional service filter e.g. 'ssh' Returns: best credential to try (decrypted), or empty if none found.
Inputs: host (required), service.
cve_to_exploitA
Given a service name/version, find matching exploits in searchsploit and Metasploit. Closes the gap between 'found a version' and 'here is the exploit to run'. service: service name e.g. 'Apache', 'OpenSSH', 'nginx', 'vsftpd', 'Samba' version: version string e.g. '2.4.49', '3.0.28', '1.24.0' banner: raw service banner (alternative to service+version โ auto-parsed) os_type: 'linux' or 'windows' (filters exploits) Returns: matching searchsploit exploits + Metasploit modules + suggested commands
Inputs: service (required), banner, os_type, version.
dig_lookupA
DNS lookup using dig. domain: target domain (e.g. 'example.com') record_type: A, AAAA, MX, NS, TXT, CNAME, SOA, ANY dns_server: optional DNS server to query (e.g. '8.8.8.8') short: return short output only
Inputs: domain (required), short, dns_server, record_type.
dig_zone_transferA
Attempt DNS zone transfer (AXFR). domain: target domain nameserver: nameserver to request transfer from
Inputs: domain (required), nameserver (required).
engagement_endA
Close the current engagement and clear scope restrictions.
Inputs: no named parameters.
engagement_findingsB
Get all findings for the current engagement. min_severity: info, low, medium, high, critical host: filter by specific host limit: max results
Inputs: host, limit, min_severity.
engagement_listA
List all engagements (past and active).
Inputs: no named parameters.
engagement_startA
Start a new engagement. Sets scope for all tools automatically. All subsequent findings, jobs, and credentials are tagged to this engagement. name: engagement name e.g. 'ClientName-WebApp-2026' scope: list of authorized targets e.g. ['192.168.1.0/24', 'example.com'] client: optional client name for reporting notes: engagement notes / rules of engagement
Inputs: name (required), scope (required), notes, client.
engagement_statusB
Show the current active engagement and its findings summary.
Inputs: no named parameters.
enum4linux_scanA
Full SMB/NetBIOS enumeration using enum4linux (-a covers all). Enumerates: users, shares, groups, password policy, OS info, printers. target: IP address of Windows/Samba host username: optional username for authenticated scan password: optional password for authenticated scan
Inputs: target (required), password, username.
extract_linksA
Extract all links (anchors, forms, scripts, images) from HTML. Useful for building a crawl tree without manually grep'ing the response body. html: HTML string base_url: optional base URL for resolving relative links only_same_origin: if True and base_url given, only return links to same host Returns: lists of links by category (anchors, forms, scripts, images, css)
Inputs: html (required), base_url, only_same_origin.
fast_port_scanA
Fast port scan using masscan (discovery) + nmap (service detection). Much faster than nmap alone for large ranges โ masscan at 5000 pps finds all open ports, then nmap does targeted -sV only on those ports. target: IP, CIDR range (e.g. '192.168.1.0/24'), or hostname ports: port range (default all ports '0-65535') rate: masscan packets per second โ higher = faster but more noisy (default 5000) Use 1000 for stealth, 10000+ for speed on local networks service_detection: run nmap -sV on discovered ports (default True) Returns: open ports with services, much faster than full nmap scan NOTE: masscan requires root/sudo for raw sockets
Inputs: target (required), rate, ports, service_detection.
ffuf_fuzzA
Fast web fuzzer using ffuf. Place FUZZ keyword in URL, headers, or POST data. url: target URL with FUZZ keyword (e.g. 'http://example.com/FUZZ') wordlist: path to wordlist (auto-selects common.txt if empty) keyword: fuzzing placeholder (default FUZZ) match_codes: show responses matching these HTTP codes filter_codes: hide responses with these HTTP codes threads: concurrent threads (default 40) data: POST data โ setting this switches to POST automatically method: HTTP method GET/POST/PUT/DELETE/PATCH headers: extra header e.g. 'Authorization: Bearer ' auto_calibrate: auto-filter similar responses to reduce noise (-ac)
Inputs: url (required), data, method, headers, +6 more.
generate_pentest_reportA
Generate a professional finding-based pentest report with attack chains and remediation.
title: report title min_severity: minimum severity to include โ info, low, medium, high, critical min_confidence: minimum confidence to include โ low, medium, high host: filter by specific host (empty = all hosts) save_to: optional file path to save the report (must be under artifacts/, /tmp, or /var/tmp) format: 'markdown' (default) or 'html' (self-contained HTML file, suitable for client delivery) confirmed_only: if True, only include findings marked 'confirmed' via update_finding_status. Use after running a validation agent for a zero-false-positive report.
Inputs: host, title, format, save_to, +3 more.
generate_reportB
Generate a structured report from completed job results. job_ids: list of job IDs to include (from list_jobs or get_job_status) title: report title format: 'markdown' or 'json'
Inputs: job_ids (required), title, format.
get_findingsA
Extract and return normalized findings from a completed job. job_id: job to extract findings from (leave empty to get all recent findings) host: filter findings by host min_severity: minimum severity to return โ info, low, medium, high, critical min_confidence: minimum confidence to return โ low, medium, high Returns: list of normalized Finding objects with host, title, severity, evidence, tool
Inputs: host, job_id, min_severity, min_confidence.
get_job_outputARead-only
Get partial output from a running or completed job. Useful for checking progress before a job finishes. job_id: job ID to read tail: number of lines from the end to return (default 100, 0 = all)
Inputs: job_id (required), tail.
get_job_statusBRead-only
Get status and result of an async job by its ID.
Inputs: job_id (required).
gobuster_dirA
Directory and file brute-force using Gobuster. url: target URL (e.g. 'http://example.com') wordlist: path to wordlist (auto-selects common.txt from dirb/seclists if empty) extensions: file extensions to search e.g. 'php,html,txt' threads: concurrent threads (default 10) exclude_codes: HTTP codes to hide (default '404'), use '' to show all follow_redirect: follow 3xx redirects
Inputs: url (required), threads, wordlist, extensions, +2 more.
gobuster_dnsB
DNS subdomain brute-force using Gobuster. domain: target domain (e.g. 'example.com') wordlist: path to wordlist (auto-selects subdomains list from seclists if empty) show_ips: show IP addresses of found subdomains
Inputs: domain (required), threads, show_ips, wordlist.
gobuster_vhostC
Virtual host discovery using Gobuster. url: base URL (e.g. 'http://example.com') wordlist: path to wordlist (auto-selects if empty) append_domain: append base domain to each vhost word
Inputs: url (required), threads, wordlist, append_domain.
html_to_textB
Strip HTML tags and extract visible text. Removes scripts, styles, tags. Returns clean readable text. html: HTML string
Inputs: html (required).
http_form_submitA
Submit an HTML form (simulates browser form POST). url: form action URL form_data: dict of form fields e.g. {"username": "admin", "password": "test"} method: POST (default) or GET headers: additional HTTP headers cookies: session cookies to include follow_redirects: follow redirects after submit (default True) Returns: response status, headers, body, redirect chain
Inputs: url (required), form_data (required), method, cookies, +2 more.
http_requestA
Make an HTTP request and inspect the full response. url: target URL method: GET, POST, PUT, DELETE, PATCH, HEAD, OPTIONS headers: dict of HTTP headers e.g. {"Authorization": "Bearer token"} cookies: dict of cookies e.g. {"session": "abc123"} data: request body for POST/PUT (string or JSON) follow_redirects: follow 3xx redirects (default True) timeout: request timeout in seconds save_to: if set, save response body to this file path (for binaries/large files) extract_text: if True and HTML, strip tags and return visible text only Returns: status, headers, body (or file path if save_to), redirect chain, timing
Inputs: url (required), data, method, cookies, +5 more.
hydra_bruteforceA
Password brute-force attack using Hydra. target: IP or hostname to attack service: protocol โ ssh, ftp, http-get, http-post-form, smb, mysql, mssql, rdp, telnet, pop3, imap, smtp, vnc username: single username to test userlist: path to username list file (overrides username) e.g. /usr/share/seclists/Usernames/top-usernames-shortlist.txt password: single password to test passlist: path to password list file (overrides password) e.g. /usr/share/wordlists/rockyou.txt port: target port (0 = use service default) tasks: parallel connections per target (default 16) stop_on_first: stop after first valid credential found (-f)
Inputs: target (required), service (required), port, tasks, +5 more.
list_artifactsA
List all files in the artifacts directory (downloads, scan outputs, payloads). Returns: filename, size, modification time per file.
Inputs: no named parameters.
list_completed_jobsA
List completed jobs, optionally filtered by tool name. tool_filter: partial tool name e.g. 'nmap', 'nikto', '' (all)
Inputs: tool_filter.
list_jobsBRead-only
List recent jobs with their statuses.
Inputs: limit.
list_unconfirmed_findingsA
List findings pending validation for the current engagement.
Returns unconfirmed findings ordered by severity so a validation agent can work through them one by one and call update_finding_status on each. host: filter by specific host (empty = all) min_severity: info, low, medium, high, critical
Inputs: host, min_severity.
msf_run_moduleA
Run a Metasploit module non-interactively via resource file. module: full module path e.g. 'exploit/windows/smb/ms17_010_eternalblue' or 'auxiliary/scanner/smb/smb_ms17_010' options: dict of module options โ RHOSTS is required and must be in scope. e.g. {"RHOSTS": "192.168.1.1", "LHOST": "192.168.1.10", "LPORT": "4444"} payload: payload e.g. 'windows/x64/meterpreter/reverse_tcp' NOTE: Non-interactive only โ sessions are auto-killed after execution.
Inputs: module (required), options (required), payload.
msf_searchA
Search Metasploit modules by keyword using a resource file (faster than -x). query: search term e.g. 'ms17-010', 'eternalblue', 'type:exploit platform:windows' Returns immediately (non-interactive).
Inputs: query (required).
msfvenom_generateA
Generate a payload using msfvenom. Saved to ~/kali-mcp/artifacts/ (chmod 600). payload: e.g. 'linux/x64/meterpreter/reverse_tcp' lhost: listener IP address lport: listener port format: elf, exe, raw, php, py, rb, sh, dll, jar, war filename: output filename (auto-generated if empty)
Inputs: lhost (required), lport (required), payload (required), format, +1 more.
nc_banner_grabA
Grab a service banner from host:port using a raw async socket. Sends a newline (or custom data) to trigger a server response. host: IP or hostname port: target port timeout: seconds to wait for response (default 10) send_data: data to send to trigger banner (default newline). Supports escape sequences like '\n', '\r\n', '\t'.
Inputs: host (required), port (required), timeout, send_data.
nc_port_checkA
Check if ports are open/closed on a host. Uses ncat (nmap-suite) if available, falls back to netcat. host: IP or hostname ports: comma/space-separated ports e.g. '80', '22,80,443' Returns immediately.
Inputs: host (required), ports (required).
nikto_scanA
Web server vulnerability scan using Nikto. target: host or URL (e.g. 'example.com' or 'http://example.com') port: target port (default 80; use 443 with ssl=True) ssl: force SSL/HTTPS max_time: nikto's internal max scan time e.g. '10m', '20m', '1h' timeout: server-side timeout in seconds (default 600 = 10 min)
Inputs: target (required), ssl, port, timeout, +1 more.
nmap_aggressive_scanB
Aggressive scan (-A): OS + version + default scripts + traceroute. targets: IPs/ranges/hostnames
Inputs: targets (required), ports.
nmap_host_discoveryA
Ping scan to discover live hosts (-sn). Fast, no port scan. targets: IPs, ranges, or CIDR (e.g. '192.168.1.0/24', '10.0.0.1-10')
Inputs: targets (required).
nmap_os_detectionB
OS detection scan (-O). Automatically uses sudo if not root. targets: IPs/ranges/hostnames
Inputs: targets (required).
nmap_port_scanA
Port scan returning job_id (async by default) or blocking until complete. targets: IPs/ranges/hostnames (space-separated for multiple) ports: '1-65535', '22,80,443', or 'top100' scan_type: 'auto' (sS if root else sT), 'sS' (SYN/root only), 'sT' (TCP connect), 'sU' (UDP), 'sA' (ACK) timing: T0-T5 (T4=fast, T3=normal, T2=polite) wait: if True, blocks until scan completes and returns full output
Inputs: targets (required), wait, ports, timing, +1 more.
nmap_service_detectionB
Detect service versions on open ports (-sV). version_intensity: 0 (light) to 9 (try all probes)
Inputs: targets (required), ports, version_intensity.
nmap_vuln_scanA
NSE vulnerability script scan. scripts: NSE script categories โ 'vuln', 'safe', 'vuln and safe', 'exploit', or specific scripts like 'smb-vuln-ms17-010'
Inputs: targets (required), ports, scripts.
nmap_xml_scanA
Port scan with XML output โ returns fully structured host/port/service data. Unlike nmap_port_scan (raw text), this parses XML into structured dicts so findings, services, and OS guesses are immediately usable without regex. targets: IPs/ranges/hostnames (space-separated) ports: '1-65535', '22,80,443', or 'top100' scan_type: 'auto' (sS if root else sT), 'sS', 'sT' timing: T0-T5 service_detection: include -sV service version probing (default True) Returns: structured dict with hosts[], each containing ports[], services, os[]
Inputs: targets (required), ports, timing, scan_type, +1 more.
nuclei_scanB
Vulnerability scanning using Nuclei templates. target: URL or IP to scan (e.g. 'http://example.com') templates: comma-separated template paths or categories e.g. 'cves/', 'vulnerabilities/', 'cves/2021/CVE-2021-44228.yaml' severity: filter by severity โ info,low,medium,high,critical (comma-separated) tags: filter by tags e.g. 'cve,rce,sqli' rate_limit: max HTTP requests per second (default 150) concurrency: concurrent template execution (default 25) auto_scan: automatic scan using wappalyzer technology detection (-as)
Inputs: target (required), tags, severity, auto_scan, +3 more.
nuclei_update_templatesB
Update Nuclei templates to the latest version.
Inputs: no named parameters.
parse_nmap_outputA
Parse nmap output from a completed job into structured data. Returns hosts, open ports, services, and OS guesses. job_id: completed nmap job ID
Inputs: job_id (required).
parse_nuclei_outputA
Parse a nuclei JSONL findings file into structured vulnerability data. findings_file: path returned by nuclei_scan as 'findings_file'
Inputs: findings_file (required).
pcap_extractA
Extract credentials and key data from a PCAP file. Extracts: HTTP requests (headers, POST data), FTP/Telnet/SMTP credentials, DNS queries, common protocol usernames/passwords. pcap_path: path to .pcap or .pcapng file Returns: structured findings by protocol
Inputs: pcap_path (required).
pcap_protocolsA
Get protocol hierarchy and conversation list for a PCAP โ overview of what's actually in the capture (HTTP, DNS, SMB, FTP, Telnet, SMTP, ICMP, ARP, etc.). Use this BEFORE pcap_extract to know what protocols to look for. pcap_path: path to .pcap or .pcapng file Returns: protocol breakdown by packet count + endpoint conversations
Inputs: pcap_path (required).
program_scope_add_targetsC
Add targets to the current program's in-scope list.
Inputs: targets (required).
program_scope_allowB
Explicitly allow a target that is otherwise out of scope.
Inputs: target (required).
program_scope_denyA
Explicitly deny a target, blocking it even if in in_scope.
Inputs: target (required).
program_scope_endA
Close the current program scope and clear scope restrictions.
Inputs: no named parameters.
program_scope_listA
List all programs (draft/active/ended).
Inputs: no named parameters.
program_scope_out_of_scopeB
Add targets to the out-of-scope (deny) list.
Inputs: targets (required).
program_scope_remove_targetsB
Remove targets from the current program's in-scope list.
Inputs: targets (required).
program_scope_startA
Start or update a named program scope.
name: program name e.g. 'ClientX-WebApp-2026' in_scope: list of allowed targets e.g. ['10.10.10.0/24', 'example.com'] out_of_scope: list of explicitly blocked targets (e.g. ['10.10.10.254']) client: optional client name rules: dict of rules of engagement: max_findings, duration_hours, reporting_format approvers: list of approved agents/humans who can modify/activate this program
Inputs: name (required), in_scope (required), rules, client, +2 more.
program_scope_statusA
Show the current active program scope.
Inputs: no named parameters.
read_fileA
Read a file from disk (artifacts, /tmp, /var/tmp, wordlists). Auto-detects file type via magic bytes. path: absolute file path max_bytes: max bytes to read (default 50KB) offset: byte offset to start from as_hex: return content as hex dump as_base64: return content as base64 (for binaries) Returns: file metadata, type detection, content (text/hex/base64)
Inputs: path (required), as_hex, offset, as_base64, +1 more.
recon_sweepA
Passive recon sweep with change detection.
Runs passive-only recon (subfinder, amass -passive, theHarvester) against a domain, records everything under a scan run, then reports what changed since the previous sweep: new subdomains, new IPs/ports, new endpoints, new vulnerabilities, and assets that disappeared.
Safe to run repeatedly/unattended (passive only). Newly-appeared assets are surfaced under 'priority_new_assets' because fresh surface is often less tested.
domain: target domain (must be within the active program scope) program: program name to group runs under (defaults to the domain) tools: subset of ['subfinder','amass','theharvester'] (default: all) tool_budget: max number of tools to run this sweep (0 = no limit)
Inputs: domain (required), tools, program, tool_budget.
scan_and_exploit_chainA
Full chain: port scan โ service detection โ CVE lookup for each service. One call that returns all exploits for all services on a target. target: IP or hostname ports: port range to scan Returns: for each open port, the service version + matching exploits
Inputs: target (required), ports.
scan_hostA
Full parallel host scan โ fires multiple tools simultaneously and waits for all. Reduces total recon time by 70% vs running tools sequentially. target: IP or hostname intensity: 'light' (quick scan), 'normal' (default), 'deep' (thorough) Returns: consolidated results from all tools with findings and next steps.
Inputs: target (required), intensity.
scan_webA
Full parallel web application scan โ nikto, gobuster, nuclei, and crawler simultaneously. depth: 'light' (quick), 'normal' (default), 'deep' (thorough with ffuf) Returns: consolidated web findings from all scanners.
Inputs: url (required), depth.
scope_addAIdempotent
Add a single target to the authorized scope. target: IP, CIDR (e.g. '192.168.1.0/24'), domain, or '*.example.com'
Inputs: target (required).
scope_clearADestructive
Clear all scope restrictions โ reverts to lab mode (all targets allowed). Use at end of engagement.
Inputs: no named parameters.
scope_listARead-only
List all authorized targets in scope. Empty list means lab mode โ all targets allowed.
Inputs: no named parameters.
scope_removeB
Remove a specific target from scope.
Inputs: target (required).
scope_setADestructive
Replace the entire scope with a new list of targets. Use this at the start of an engagement to set all authorized targets at once. targets: list of IPs, CIDRs, domains e.g. ['192.168.1.0/24', 'example.com', '*.example.com']
Inputs: targets (required).
screenshot_urlA
Take a screenshot of a single web URL using gowitness. Useful for quick visual triage of login panels, admin interfaces, etc. url: target URL (e.g. 'http://example.com/admin') timeout: per-request timeout in seconds Returns: path to saved screenshot PNG file
Inputs: url (required), timeout.
screenshot_urlsA
Take screenshots of multiple URLs using gowitness. Use after gobuster/crawl to visually triage all discovered endpoints at once. urls: list of URLs to screenshot threads: concurrent screenshot workers (default 4) timeout: per-request timeout in seconds Returns: directory containing all PNG screenshots + JSONL results
Inputs: urls (required), threads, timeout.
searchsploit_get_pathA
Get the full filesystem path to an exploit by its EDB-ID. edb_id: Exploit-DB ID number (e.g. '39446')
Inputs: edb_id (required).
searchsploit_searchA
Search Exploit-DB local database using searchsploit. query: search terms (e.g. 'apache 2.4', 'wordpress 5.0', 'ssh') exact: exact & order match on title (stricter) title_only: search only exploit titles cve: search by CVE ID (e.g. '2021-44228') โ overrides query exclude: exclude results matching this pattern (pipe-separated: 'dos|PoC') Returns results immediately (local DB search).
Inputs: cve, exact, query, exclude, +1 more.
server_healthA
Preflight check โ verifies all tool binaries, Python deps, and wordlists. Run this at the start of an engagement to avoid surprises mid-workflow. Returns: status per category, missing tools with install hints, available wordlists.
Inputs: no named parameters.
sqlmap_scanA
Automated SQL injection detection and exploitation using sqlmap. url: target URL with parameter (e.g. 'http://site.com/page.php?id=1') data: POST data string (e.g. 'user=foo&pass=bar') level: test level 1-5 (higher = more tests, slower) risk: risk level 1-3 (higher = more aggressive, may alter data) dbms: force DBMS type e.g. 'mysql', 'postgresql', 'mssql', 'oracle' technique: SQL injection techniques โ B=Boolean, E=Error, U=Union, S=Stacked, T=Time, Q=Inline (default all: BEUSTQ) enumerate_dbs: enumerate databases (--dbs) enumerate_tables: enumerate tables (--tables) dump: dump table data โ use with database and table params database: target database name (-D) table: target table name (-T) cookie: HTTP cookie header value random_agent: use random User-Agent NOTE: Always runs non-interactively (--batch).
Inputs: url (required), data, dbms, dump, +9 more.
ssh_enum_privescA
Enumerate privilege escalation vectors on a Linux host via SSH. Runs checks for: SUID binaries, capabilities, sudo permissions, cron jobs, world-writable files, kernel version, /etc/passwd readables, PATH/LD env vars. host: target IP or hostname username: SSH username password: SSH password key_file: SSH private key path port: SSH port (default 22) Returns: structured findings by category, OR a clear error if SSH failed.
Inputs: host (required), username (required), port, key_file, +1 more.
ssh_execA
Execute a single command via SSH (uses paramiko, no sshpass dependency). host: target IP or hostname username: SSH username password: SSH password (leave empty if using key) key_file: path to private key file (leave empty if using password) command: shell command to execute port: SSH port (default 22) timeout: command timeout in seconds (default 60) Returns: stdout, stderr, return_code (or error if connection/auth failed)
Inputs: host (required), username (required), port, command, +3 more.
subfinder_enumerateA
Passive subdomain enumeration using subfinder. domain: target domain (e.g. 'example.com') all_sources: use all available sources (slower but thorough) threads: concurrent goroutines for resolving (default 10) output_json: return JSONL output for structured parsing
Inputs: domain (required), threads, all_sources, output_json.
theharvester_searchA
OSINT gathering using theHarvester (emails, subdomains, IPs, URLs). domain: target domain or company name source: comma-separated data sources. Defaults to sources that work without API keys. Full list: google, bing, linkedin, github, dnsdumpster, crtsh, hackertarget, otx, rapiddns, shodan (needs key), sublist3r, threatminer, urlscan, duckduckgo Use 'all' only if you have API keys configured. limit: max search results (default 500) dns_resolve: perform DNS resolution on discovered subdomains
Inputs: domain (required), limit, source, dns_resolve.
tshark_queryA
Run an arbitrary tshark query on a PCAP file. pcap_path: path to PCAP file display_filter: Wireshark display filter e.g. 'http.request', 'tcp.port==21', 'smb', 'dns.qry.name', 'frame.number == 42' fields: comma-separated tshark fields to extract e.g. 'ip.src,ip.dst,tcp.port,http.request.uri' max_lines: max output lines (default 200)
Inputs: pcap_path (required), fields, max_lines, display_filter.
update_finding_statusA
Update the validation status of a finding.
Called by a validation agent after manually verifying a finding. finding_id: the finding's id (from list_unconfirmed_findings) status: 'confirmed' โ finding is real and exploitable 'false_positive' โ finding is not real, exclude from report 'unconfirmed' โ reset back to pending (if re-verification needed)
Inputs: status (required), finding_id (required).
web_crawlA
Crawl a web application by following links organically. Discovers endpoints, forms, and interesting paths that wordlist brute-force misses. url: starting URL (e.g. 'http://example.com/') max_depth: how many link levels deep to follow (default 2) max_pages: max pages to visit (default 50) include_external: also collect external links (not crawled, just listed) timeout: per-request timeout in seconds Returns: discovered URLs by category, forms, and interesting endpoints
Inputs: url (required), timeout, max_depth, max_pages, +1 more.
whois_lookupA
WHOIS lookup for a domain or IP address. target: domain (e.g. 'example.com') or IP address
Inputs: target (required).
wpscan_scanA
WordPress vulnerability scanner using WPScan. url: WordPress site URL (e.g. 'http://example.com') enumerate: what to enumerate โ comma-separated: vp=vulnerable plugins, vt=vulnerable themes, u=users, p=all plugins, t=all themes, tt=timthumbs, cb=config backups, dbe=db exports, m=media api_token: WPScan API token for vulnerability data (get free at wpscan.com) detection_mode: mixed (default), passive, aggressive random_user_agent: randomize user agent per request disable_tls_checks: disable SSL certificate verification throttle_ms: milliseconds between requests (0 = no throttle)
Inputs: url (required), api_token, enumerate, throttle_ms, +3 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogโฆ
TDQS
Scored across 101 tools
Many tools have overlapping purposes: multiple port scanners (nmap_port_scan, nmap_xml_scan, fast_port_scan), multiple subdomain tools (subfinder_enumerate, amass_enum, theharvester_search, recon_sweep), and two parallel scope management systems (scope_* and program_scope_*). An agent would frequently struggle to pick the correct tool without reading full details.
There are two main conventions: category_action (nmap_port_scan, scope_add, asset_mark_host) and verb_noun (get_job_status, list_artifacts). Some names like server_health, html_to_text, and pcap_protocols break the pattern entirely. It's readable but not uniform.
101 tools is far beyond reasonable for a single MCP server. The calibration specifies 50+ as an extreme mismatch, and this server has more than double that. The sheer number overwhelms tool selection and creates unnecessary redundancy.
The toolset covers the entire penetration testing lifecycle: recon (nmap, subfinder, theharvester), scanning (nikto, nuclei, gobuster), exploitation (sqlmap, hydra, metasploit), post-exploitation (ssh_exec, creds_store), and reporting (generate_pentest_report, analyze_findings). No obvious gaps for the Kali Linux domain.
Related MCP Connectors
AI pentesting: run scans, triage vulnerabilities, review PRs, manage schedules and assets.
Query your team's drift, vulnerability, and upgrade data from any AI assistant. OAuth 2.1, 51 tools.
Pay-per-call cybersecurity for AI agents: vuln scans, threat intel, compliance, code security.
Security scanner for AI-written code: SAST, secrets, dependencies, IaC, containers and DAST.
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceEnables AI assistants to execute penetration testing commands and security tools on Kali Linux remotely. Supports automated reconnaissance, vulnerability scanning, and CTF solving through integration with 25+ offensive security tools like nmap, gobuster, and nuclei.16-
- FlicenseNot gradedqualityDmaintenanceConnects AI assistants to 55+ Kali Linux security tools for automated CTF solving, penetration testing, and security analysis across 7 categories including cryptography, forensics, web security, and binary exploitation.80-
- AlicenseNot gradedqualityDmaintenanceEnables AI assistants to perform penetration testing and security assessments by exposing 60+ Kali Linux security tools including network scanning, web security testing, password cracking, exploitation frameworks, and OSINT capabilities through an AI-friendly interface.2MIT
- AlicenseAqualityDmaintenanceEnables AI assistants to perform authorized penetration testing and security assessments by exposing 20+ Kali Linux security tools (nmap, sqlmap, gobuster, hydra, etc.) through a safe, validated interface with command allowlists, rate limiting, and input sanitization.191MIT