aegis_evaluate_proposalLABSCO SUMMARY
A structured answer to should I do this.
The value is the shape of the output: not advice in prose but a decision, a confidence and a rationale, kept in a hash-chained trail — which is what makes it readable months later by someone asking why the agent proceeded.
What it is
A governance gate an agent can call on itself. A proposal is scored on six dimensions — Risk, Profit, Novelty, Complexity, Quality, Utility — and comes back as a structured decision with confidence and rationale.
What you get
- A proposal evaluated into PROCEED, PAUSE, HALT or ESCALATE, with confidence scores and rationale
- A fast risk check for when a full evaluation is more than the moment needs
- The thresholds a decision is being measured against
- The scoring guide, so the six dimensions are applied the same way each time
- A hash-chained audit trail behind the decisions
- An audit record written against NIST AI RMF and EU AI Act Annex IV
Requirements
A recent Python and pip. It runs locally in sandbox mode with no signup; an Aegis API key in the server's environment unlocks decision history, usage reports and risk checks, and a hosted streamable-HTTP endpoint carries the wider tool surface.
Setup effort
One command — pip install "aegis-governance[mcp]"
USE CASES
01
Stopping an agent before an irreversible step
The proposal is scored and returned as a decision the agent can act on, rather than as commentary it can talk itself out of.
02
Screening quickly when the full gate is too slow
A risk check for the many small decisions that do not warrant a six-dimension evaluation.
03
Understanding why a proposal was held
Read the thresholds it was measured against and the guide behind the scores, instead of arguing with the verdict.
04
Producing a record compliance can read
Each decision leaves a hash-chained entry with its rationale, which is the artefact an audit actually asks for.
05
Scoring two proposals the same way
The scoring guide makes the dimensions consistent, so this week's PROCEED means what last week's did.
10 tools
Listed in project docs. Your client may show a different list.
aegis_quick_risk_checkaegis_check_thresholdsaegis_get_scoring_guideaegis_record_proposalaegis_list_proposalsaegis_verify_proposalsaegis_list_decisionsaegis_get_decisionaegis_crypto_statusBEFORE YOU INSTALL
What to know first
Tools you get
10 tools named in the project's documentation.
Requires
aegis-mcp-server on PATH
Optional
AEGIS_API_KEY in your environment
INSTALLATION
Connect your client
Server setup
pip
TERMINAL
pip install "aegis-governance[mcp]"Hosted endpoint
ENDPOINT URL
https://mcp.aegis.undercurrentholdings.com/mcpAdd this hosted endpoint to your MCP client.
MCP Clients
Claude Code
CLAUDE CODE · TERMINAL
claude mcp add aegis -- aegis-mcp-serverThe local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
Cursor, Windsurf
JSON
{
"mcpServers": {
"aegis": {
"command": "aegis-mcp-server"
}
}
}The local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
VS Code
VS CODE · JSON
{
"servers": {
"aegis": {
"type": "stdio",
"command": "aegis-mcp-server"
}
}
}The local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
Claude Code — remote
CLAUDE CODE — REMOTE · TERMINAL
claude mcp add --transport http aegis https://mcp.aegis.undercurrentholdings.com/mcp \
--header "Authorization: Bearer <YOUR_API_KEY>"The local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
Cursor — remote
CURSOR — REMOTE · JSON
{
"mcpServers": {
"aegis": {
"type": "streamable-http",
"url": "https://mcp.aegis.undercurrentholdings.com/mcp",
"headers": {
"Authorization": "Bearer <YOUR_API_KEY>"
}
}
}
}The local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
VS Code — remote
VS CODE — REMOTE · JSON
{
"servers": {
"aegis": {
"type": "http",
"url": "https://mcp.aegis.undercurrentholdings.com/mcp",
"headers": {
"Authorization": "Bearer <YOUR_API_KEY>"
}
}
}
}The local server runs in sandbox mode without a key. To unlock decision history, usage reports and risk checks, get a free key at portal.undercurrentholdings.com and set AEGIS_API_KEY in the server's environment — in a JSON client entry, an env block beside "command". The remote configurations send the same key as Authorization: Bearer <YOUR_API_KEY> instead.
