Are you the author? Sign in to claim
MCPhantom is a local web dashboard for MCP endpoint recon. Discover resources, templates, tools, and prompts, then inter
MCPhantom is a local web dashboard for MCP endpoint reconnaissance and exploitation. Point it at any MCP URL, discover exposed resources, templates, tools, and prompts, then interact with them from a clean security console UI. Runs locally at 127.0.0.1:1337; no cloud, no setup beyond Python. Project's Documentation can be found here.
[!WARNING] MCPhantom is an extensible security auditing framework; not a turnkey, universal scanner. It provides a solid foundation for MCP focused reconnaissance and vulnerability testing, but it is intentionally designed as a starting skeleton that you are expected to adapt to your targets, environments, and methodology.
- Not guaranteed to work out of the box on every MCP server : Capabilities, schemas, transports, and response formats vary widely across implementations. Classification, payload delivery, and proof detection may need tuning per target.
- Payload coverage is deliberately limited : MCPhantom includes representative probes for classes such as SQL injection, command injection, SSRF, path traversal, IDOR, and information disclosure, but it does not ship exhaustive wordlists or engine scale fuzzing comparable to tools like sqlmap, Burp Intruder, or commercial DAST platforms.
- Proof heuristics are best-effort : Findings are scored from response signals (errors, data leaks, version strings, reflected output, etc.). False positives and false negatives are possible without target-specific customization. This is exactly why the framework also allows you to do manual auditing.
- You are encouraged to extend it : The project is open source so you can grow payload libraries, add plugins, refine classifiers, integrate with your CI/CD pipeline, or harden detection logic for your use cases.
- The AutoPwn feature is a WIP and it is highly discouraged to use it in live production encironments. Only use this feature on CTFs and/or local labs.
https://github.com/user-attachments/assets/b62d3a80-2b38-4b5c-a11b-029192279470
| Category | Feature | Description |
|---|---|---|
| Discovery | MCP enumeration | Discovers resources, resource templates, tools, and prompts from a target MCP endpoint |
| Capability classification | Tags capabilities (e.g. URL, command, database, path, ID) to decide which tests apply | |
| Interface | Web dashboard | Browser-based console at http://127.0.0.1:1337 with dark/light themes |
| Live interaction | Read resources/templates, invoke tools, run prompts, and fill template placeholders from the UI | |
| AutoPwn | Smart targeted scanning | Runs plugin-matched checks only where tags/parameters suggest relevance - avoids blind fuzzing everywhere |
| Parallel execution | Runs multiple capability checks concurrently with per-check and per-payload timeouts | |
| Streaming results | NDJSON progress stream with live coverage, findings, and completion percentage | |
| Scan coverage report | Per-capability status (vulnerable, clean, timeout, error) with hit counts and notes | |
| Plugins | SQL injection | Single-column UNION probes, DB version extraction, and schema/table enumeration (SQLite, MySQL, PostgreSQL, MSSQL, Oracle) |
| Command injection | Shell metacharacter and allowlist-bypass style payloads with MCP response parsing | |
| SSRF | Internal/localhost and metadata-style URL probes | |
| Path traversal | Common filesystem path escape payloads | |
| IDOR | Identifier manipulation probes on ID-like parameters | |
| Information disclosure | Baseline and verbose/error-triggering requests for sensitive output | |
| Engine | Central payload repository | All payloads live in audit/payloads.py - add categories and entries without rewriting plugins |
| Proof scoring (SQLi) | Ranks findings by evidence strength; filters reflection-only and duplicate error responses | |
| Plugin architecture | Modular plugins loaded via audit/plugin_loader.py - straightforward to add new vulnerability classes | |
| Extensibility | Open source | Fork, extend payload lists, add plugins, and tailor detection to your targets and workflows |
python web_server.py
# Open http://127.0.0.1:1337, enter an MCP endpoint, run Start Recon.
[!CAUTION] MCPhantom is intended solely for authorized security assessments, research, and educational purposes. Use it only against systems you own or have explicit permission to test. See the LICENSE file for warranty and liability terms.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
@microsoft✓ OfficialBrowser automation using accessibility snapshots instead of screenshots
@googleapis✓ OfficialGoogle's universal MCP server supporting PostgreSQL, MySQL, MongoDB, Redis, and 10+ databases
@github✓ OfficialOfficial GitHub integration for repos, issues, PRs, and CI/CD workflows