Aegis
If you're building agents that do anything consequential, this gives them a decision checkpoint they can call before acting. It runs six quantitative gates (risk, profit, novelty, complexity, quality, utility) and returns PROCEED, PAUSE, HALT, or ESCALATE with scored rationale. The real win is the hash-chained audit trail: every decision gets a tamper-evident log entry you can hand to compliance. Works locally in sandbox mode with no signup, or connect to the hosted server for full history and verification tools. Designed for NIST AI RMF and EU AI Act artifact generation, so it's aimed at regulated environments where "the agent did it" isn't documentation.
- tools
- 6
- auth
- Secret needed
- transport
- STDIO
- env var
- 1
- last updated
- Sep 18, 2026
Tools
6 toolsPublic tool metadata for what this MCP can expose to an agent.
aegis_evaluate_proposalEvaluate an engineering proposal through the AEGIS gate checks.
Evaluate an engineering proposal through the AEGIS gate checks.
No parameter schema in public metadata yet.
aegis_quick_risk_checkQuick risk threshold check without a full Bayesian evaluation.
Quick risk threshold check without a full Bayesian evaluation.
No parameter schema in public metadata yet.
aegis_healthCheck AEGIS API health (service status, version, components).
Check AEGIS API health (service status, version, components).
No parameter schema in public metadata yet.
aegis_list_decisionsList recent governance decisions for the authenticated customer.
List recent governance decisions for the authenticated customer.
No parameter schema in public metadata yet.
aegis_get_decisionFetch a specific governance decision by ID, with full gate detail.
Fetch a specific governance decision by ID, with full gate detail.
No parameter schema in public metadata yet.
aegis_get_usageGet the authenticated customer's evaluation usage report.
Get the authenticated customer's evaluation usage report.
No parameter schema in public metadata yet.
Sponsors
Agent CaffeineSponsoredClose your lid, come back to finished work
SnugSponsoredThe clipboard manager for the AI eraPromote your product hereReach 125k+ visitors monthly
WaxSponsoredBlock anything with supervision, without erasing your iPhonePromote your product hereReach 125k+ visitors monthly
knobsSponsoredSee your site on any phone, any day, in any languageConfiguration
AEGIS_API_KEYsecretAPI key for the hosted AEGIS API. The hosted service is offline, so tool calls do not return evaluations. See https://undercurrentholdings.com/status/
Readme
AEGIS Governance — MCP Server
Status: the hosted AEGIS service is offline (since September 2026). The hosted API and the hosted MCP endpoint (
mcp.aegis.undercurrentholdings.com) are offline. The local server does not evaluate proposals itself, so no tool returns an evaluation at this time. Current status: https://undercurrentholdings.com/status/
What the package is
aegis-governance on PyPI is a Python client for the AEGIS API. It also contains a local
MCP server. The package installs one console script, aegis-mcp-server. This script starts
a local MCP server that uses the stdio transport. The package has no other command-line tool.
The local server has no evaluation engine. By default, the client sends all requests to the hosted AEGIS API. Sandbox mode (the mode without an API key) also uses the hosted API. While the hosted service is offline, tool calls do not return evaluations.
Install
pip install "aegis-governance[mcp]"
The [mcp] extra is necessary for aegis-mcp-server. On Python < 3.10, the extra installs
nothing, and aegis-mcp-server exits with an install hint.
Local stdio MCP configuration
Claude Code
Details
- Repository
- github.com/undercurrentai/aegis-mcp
- Website
- aegis.undercurrentholdings.com
- Package
aegis-governance- Transport
- STDIO
- Auth
- Required
- Registry
- MCP registry, active