Purpose-built forensic MCP backend for Protocol SIFT. Exposes 43+ typed forensic tools through the Model Context Protocol (MCP) using stdio transport. Designed for use with Claude Code as the primary agentic execution engine on SANS SIFT Workstation. Features cross-artifact contradiction detection, evidence-triggered self-correction, and read-only enforcement at the transport layer.
Critical failures across all dimensions. Tool definitions exist in tool_catalog.py but lack visible input schemas, parameter descriptions, and output documentation. 60 tools declared with only brief descriptions (10-50 chars); no evidence of JSON Schema definitions, type constraints, or parameter documentation. No error handling guidance. Naming is inconsistent (verb_noun mostly present but some tools like 'classify_missing_artifact' are vague). This server exposes raw forensic tool wrappers without the abstraction layer required for LLM-safe composition.
Run comprehensive analysis on collected artifacts
Compare disk and memory artifacts to detect discrepancies and cross-artifact contradictions
Find temporal clusters in timeline data for event correlation
Flag and record discrepancies between disk and memory findings
Analyze Volume Shadow Copies for historical artifact recovery
Extract Program Compatibility Assistant cache for execution history
Extract Shimcache registry entries for application execution history
NO INPUT SCHEMAS VISIBLE: 60 tools declared with descriptions only; no JSON Schema definitions, parameter types, or constraints found in source. Cannot verify parameter validation, type safety, or LLM-safe composition.
DESCRIPTIONS TOO BRIEF: Most tool descriptions 10-50 characters (baseline 194 chars). Descriptions like 'Verify the integrity of forensic evidence files' lack context for LLM selection, prerequisites, or when to use vs similar tools.
NO PARAMETER DOCUMENTATION: Tools like state.read_state show input schema with case_id but no descriptions for parameters. LLMs cannot infer what case_id format is expected (UUID? string? path?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
Extract System Resource Usage Monitor data for network and application activity
Hunt for suspicious patterns using Hayabusa rules
Query and analyze Sigma rule detection results
Hunt for suspicious patterns using Sigma rules
Scan artifacts against Sigma rules for detection
Classify and analyze missing or unavailable artifacts
Detect the triage layout of a disk image
Extract browser history from multiple browsers
Extract Windows Jump Lists for recent file access analysis
Extract Windows .lnk (shortcut) files for user activity analysis
Parse the NTFS Master File Table with MFTECmd. Produces a CSV timeline of all MFT entries including standard information and filename attribute timestamps.
Extract PowerShell command history for execution analysis
Extract Windows Prefetch files with PECmd. Extracts execution timestamps, run counts, and referenced file/directory paths from .pf files.
Extract Windows Recycle Bin metadata for deleted file recovery
Extract Windows Registry file access artifacts
Extract Windows Registry Run keys for persistence analysis
Extract Windows Scheduled Tasks for persistence and execution analysis
Extract Windows Shellbags registry entries for user activity analysis
Parse an extracted $UsnJrnl:$J stream with MFTECmd. Extracts NTFS USN Journal entries for file system activity timeline.
Extract comprehensive Windows artifacts including event logs, registry, and file system artifacts
Parse the Amcache.hve registry hive with AmcacheParser. Extracts SHA-1 hashes and metadata for executables that have been run on the system.
List deleted files from the file system
Summarize Windows Event Log (.evtx) files
Get the provenance and chain of custody information for evidence
Verify the integrity of forensic evidence files
Build an index of all investigation reports
Generate a graph representation of findings and correlations
Merge graphs from multiple hosts for cross-host analysis
Serve the generated graph for visualization
Add a new finding to the case state
Generate a coverage report of artifacts analyzed
Verify environment prerequisites before investigation begins
Generate the final forensic investigation report
Initialize a new forensic investigation case
Detect process injection techniques in memory
Detect the memory profile and volatility version for memory analysis
List loaded DLLs from memory dump
List running processes from memory dump
Scan network connections from memory dump
Scan processes in memory for suspicious activity
Generate NTFS mount command arguments
Load and prepare memory dump for analysis
Describe the available tool catalog and tool domains
Export the full execution trace as a list of audit entries. Reads every entry from audit.jsonl and returns entries in chronological order. Each entry captures a started or completed event for a tool invocation, including command line, parameters, exit code, duration, and any CORRECTION_EVENT produced.
Get a specific finding by ID from case state
Get all findings from case state
Get investigation gates and decision points
Return the current authoritative case state summary. Loads the state.json file and returns a compact summary with current finding/execution counts, status, open questions, and the 10 most recently added findings.
Record an analysis lane execution in case state
Build a comprehensive timeline from multiple artifact sources
Query the built timeline for specific events and time ranges
Scan files on disk with YARA rules
Scan memory dump with YARA rules
VAGUE TOOL NAMES: 'classify_missing_artifact', 'run_analysis', 'handle_*' lack action clarity. LLMs cannot distinguish when to call vs similar tools. Rename to 'categorize_missing_artifact', 'execute_comprehensive_analysis'.
NO ERROR HANDLING GUIDANCE: No evidence of error categorization (retryable vs fatal), recovery hints, or actionable error messages. Agents cannot self-correct on failures.