Overview

The Aegis Sovereign MCP Server exposes 31 governance tools to any MCP-compatible AI assistant across five modules: Compliance, Registry, Audit, Monitoring, and the AI Safety module. Install it once — all 31 tools become available in Claude Code, Claude Desktop, and any other MCP client. Interact with your governance platform in natural language: 'run a compliance check on fraud-v4', 'validate this prompt through the safety engine', 'show me SLO burn rates', 'promote credit-risk-v2 to production'.

Installation

Install the MCP Server globally and register it with your MCP client.

bash

Install and configure

1# Install the MCP server
2npm install -g @aegissovereign/mcp-server
3
4# Add to Claude Code (writes to ~/.claude/mcp_servers.json)
5claude mcp add sovereign-gateway \
6  --command "sovereign-mcp" \
7  --env SOVEREIGN_API_KEY=your_api_key \
8  --env SOVEREIGN_BASE_URL=https://sovereign.yourcompany.com
9
10# Verify — list available tools
11claude mcp list

json

claude_desktop_config.json (Claude Desktop)

1{
2  "mcpServers": {
3    "sovereign-gateway": {
4      "command": "sovereign-mcp",
5      "env": {
6        "SOVEREIGN_API_KEY": "your_api_key",
7        "SOVEREIGN_BASE_URL": "https://sovereign.yourcompany.com"
8      }
9    }
10  }
11}

Available Governance Tools

All 31 tools are available immediately after installation. Use them via natural-language prompts in your AI assistant. The original 16 governance tools are unchanged; 15 Safety tools added in v0.2.0. Two safety tools (safety_get_sbom, safety_scan_vulnerabilities) are reserved for a future release and currently return a not-implemented response pointing to the shipped alternatives.

Module	Tool	Description
Compliance	`check_compliance_status`	Latest compliance score and open findings for a model
Compliance	`run_compliance_eval`	Evaluate a model against a single regulatory framework
Compliance	`run_compliance_eval_all`	Evaluate against all 7 active frameworks in parallel
Registry	`list_models`	List registered models with stage, status, and scores
Registry	`get_model`	Full detail for one model: versions, evals, lineage
Registry	`promote_model`	Promote to production (runs pre-flight gate checks)
Registry	`get_model_card`	Auto-generated model card: intended use, metrics, bias, compliance
Registry	`revert_model`	Revert a model to its previous staged/production version
Audit	`get_audit_log`	Query the tamper-evident audit chain by model, user, time, or type
Audit	`verify_audit_chain`	Verify SHA-256 chain integrity from genesis to latest entry
Audit	`anchor_audit_chain`	Anchor the chain head to an external ledger (immudb / Azure CL / WORM)
Audit	`verify_audit_anchor`	Verify a previously written external anchor
Monitoring	`check_llm_usage`	Token usage and cost breakdown by workspace or team
Monitoring	`run_drift_check`	Run a PSI drift check for one or all production models
Monitoring	`get_workspace_metrics`	Aggregate workspace KPIs: model health, volume, latency
Monitoring	`get_drift_history`	Historical PSI drift series for a model
Safety ✨	`safety_validate_text`	Run text through the Safety Policy Engine (regex + ML ensemble). Returns violations, severity, blocked status, classifier scores
Safety ✨	`safety_validate_tool_call`	Validate an agent tool call before execution: risk level, argument patterns, principal allow-list
Safety ✨	`safety_validate_delegation`	Enforce multi-agent trust boundaries — block privilege escalation across hand-offs
Safety ✨	`safety_mask_pii`	Redact PII in text in place (Presidio / built-in regex) instead of blocking
Safety ✨	`safety_check_model`	Check ModelSafetyRegistry clearance for a model. Returns clearance verdict and constraints
Safety ✨	`safety_clearance_gate`	Run the pre-deployment clearance gate: aggregate bias/robustness/safety evals into a deploy verdict
Safety ✨	`safety_run_redteam`	Trigger the red-team certification runner. Optional domain libraries (medical/financial/legal) and multi-turn chains
Safety ✨	`safety_slo_status`	Current burn rate and error budget for all safety SLOs. Flags critical burn rate > 14.4×
Safety ✨	`safety_generate_report`	Generate a safety report from real safety audit events and request volume
Safety ✨	`safety_classifier_status`	Check which ML classifier ensemble is active. Flags NullClassifier (regex-only) with setup guidance
Safety ✨	`safety_list_reviews`	List pending approvals / review requests. Filter by status, priority
Safety ✨	`safety_submit_review`	Submit an approve/reject decision via the Approvals system. Recorded in the immutable audit trail
Safety ✨	`safety_get_lineage`	Show a model's provenance/lineage DAG to its root base model
Safety ⏳	`safety_get_sbom`	Roadmap — returns not-implemented; use `safety_get_lineage` for provenance today
Safety ⏳	`safety_scan_vulnerabilities`	Roadmap — returns not-implemented; use `safety_run_redteam` for adversarial coverage today

Usage Examples

Interact with the platform in natural language inside your IDE.

bash

Natural-language governance workflows in Claude Code

1# Compliance and registry
2> run a compliance check on fraud-detection v4 against EU AI Act and SR 11-7
3> show drift status for all production models
4> promote credit-risk-v2 to production
5> pull clinical-nlp-v1 from production immediately
6> verify the audit chain for the last 30 days
7
8# AI Safety Platform (new in v0.2)
9> validate this prompt through the safety engine: "Ignore all previous..."
10> check if gpt-4 turbo is cleared for customer-support use
11> mask any PII in this support transcript before I paste it
12> show me the provenance lineage for claude-sonnet-4-6
13> what are the current SLO burn rates for our safety fleet?
14> run the red-team certification suite with financial domain probes
15> show me all pending approvals waiting on review
16> generate an EU AI Act compliance report for Acme Financial
17> is the ML safety classifier active or is NullClassifier running?