Loading ToolBench
Multi-capability MCP server repository containing attack surface management, BloodHound integration, .NET reversing, web security, network operations, and other security analysis tools
Critical failures across naming, descriptions, and schemas. 30 of 42 tools (71%) have NO visible input schemas in the provided source code. Most tool descriptions are present but generic/minimal (avg ~80 chars, well below 194-char baseline). Tool names lack consistent verb-noun structure (e.g., 'dotnet_scan_binaries' vs 'scan_binaries'). No parameter descriptions visible for most dotnet tools. Output schemas completely undocumented. Error handling absent. Security concerns: exif_write and exif_copy enable arbitrary metadata injection with no validation or audit trail.
Authenticate to BloodHound CE and Neo4j. Call once at the start of a session, or when credentials change.
Decompile .NET methods
Decompile a .NET module
Decompile a .NET type
Download a NuGet package
Get call flows to a specific method
List all methods in .NET assemblies
List methods within a specific type
30 of 42 tools (71%) have NO visible input schemas. BloodHound and dotnet tools show empty {} or missing schema definitions entirely.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
List namespaces in .NET assemblies
List all types in .NET assemblies
List types within a specific namespace
Scan .NET binaries for analysis
Search .NET types and methods by name
Search for references in .NET assemblies
Copy all metadata from one file to another. Transfers EXIF/XMP/IPTC tags between files. Use to transplant payloads from a crafted file into a target-compatible format, or to replicate metadata from a legitimate file into a malicious one.
Read EXIF/XMP/IPTC metadata from a file. Returns all metadata tags by default, or specific tags if provided. Useful for inspecting uploaded files, checking if metadata survives server-side processing, or examining downloaded images for information leakage.
Strip all metadata from a file. Removes every EXIF/XMP/IPTC tag. Use to create a clean baseline file before injecting specific payloads, or to test whether a server-side upload handler strips metadata by comparing before/after.
Write metadata tags to a file. Injects arbitrary values into EXIF/XMP/IPTC fields. Common security payloads: XSS in Comment/Artist/Copyright/ImageDescription, SSRF URLs in GPSImgDirection or XMP fields, command injection strings in metadata that gets logged or processed by backend tools.
Find all users and computers that are members of the Domain Admins group
Find all Kerberoastable users
Find users vulnerable to AS-REP roasting (no pre-auth required)
Find computers without SMB signing enabled
Find computers running the WebClient service
Find principals with DCSync privileges
Find domain users with local admin rights
Find certificate templates vulnerable to ESC1
Find CAs vulnerable to ESC8 (NTLM relay to HTTP enrollment)
Find Azure/Entra Global Administrator role members
Find tier zero users vulnerable to Kerberoasting
Find NTLM relay attack opportunities
Find paths from Domain Users group to tier zero
Find paths from Entra principals to on-prem tier zero
Find paths from Kerberoastable users to Domain Admins
Find attack paths from owned principals
Map the PKI certificate authority hierarchy
Find shortest attack paths from any principal to tier zero assets
Find paths to computers with unconstrained delegation
Find where tier zero principals are logged in
Find computers running unsupported operating systems
Find enabled users whose password hasn't been changed in over a year
Map all domain trust relationships
Run a Cypher query against the BloodHound Neo4j database.
Output schemas completely undocumented. No tool declares what fields it returns, data types, or structure. LLMs cannot plan downstream calls or extract required IDs. Violates pattern:tool and pattern:response-shaper.
exif_write and exif_copy enable arbitrary metadata injection (XSS, SSRF, command injection payloads) with NO input validation, sanitization, or audit trail. Descriptions acknowledge security payloads but tool provides no guards. Violates pattern:tool-gateway and pattern:audit-trail.
connect() accepts credentials (username, password, neo4j_password) as tool parameters. Use server-side secret injection.' Secrets in params leak into logs and prompt history. Violates pattern:secret-injection.
No error handling guidance. Tools provide no recovery hints, retryability classification, or actionable error messages. LLMs cannot self-correct on failures. Violates pattern:recovery-guide and pattern:error-classification.