Aegis Guardrails
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AEGIS_CLIENT | No | Identifies the coding agent host the MCP server was registered for; `aegis init` sets it at registration time (e.g. `codex mcp add aegis --env AEGIS_CLIENT=codex-cli -- ~/.local/bin/aegis-mcp` or `claude mcp add --scope user aegis -e AEGIS_CLIENT=claude-code -- ~/.local/bin/aegis-mcp`). | |
| AEGIS_MAX_JIT_TOKENS | No | Token budget for the rules returned by `aegis_scan` (default 2,000 tokens); equivalent to the `jit.max_tokens` config setting. | 2000 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| aegis_scanA | Call before editing files: returns the governance rules that apply to them, ranked by ledger fitness and clamped to the JIT token budget. Mandatory rules (gates, invariants) take at most two slots; the others are listed in omitted_mandatory. Gates and invariants are served from the reviewed base commit (rules_base); branch-local edits to them never bind and are listed in unreviewed_rules. stale_rules lists rules edited since the last aegis sync. If any rule file cannot be loaded the scan fails (status FAIL) and names it. |
| aegis_report_outcomeA | Report whether the rules you were shown helped. The server ties the report to the tree it computes now: with an |
| aegis_create_ruleA | Turn a natural language concern into a prompt that drafts a governance rule. It writes nothing: save the drafted rule under .aegis/rules/ yourself. |
| aegis_list_rulesA | List the governance rules as {rules, stale_rules, ignored}: every rule with its type, hypothesis and ledger fitness (a file that cannot be loaded is listed with _error), the rules edited since the last aegis sync, and markdown under .aegis/rules that is not a rule. |
| aegis_capture_insightA | Record a user correction as an unconfirmed suggestion in user scope. Agent suggestions never count toward proposals or debits and never appear in aegis_scan; a suggestion upgrades to W2 only when its quote matches a hook-captured prompt in the same session. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 5 tools
Each tool has a clearly distinct purpose: capture_insight records corrections, create_rule drafts a rule, list_rules enumerates rules, report_outcome provides feedback, and scan retrieves applicable rules. No two tools perform the same action, so an agent can easily select the right one.
All tools share the aegis_ prefix and follow a verb_noun pattern except aegis_scan, which is verb-only. This minor deviation keeps the set readable and predictable.
Five tools is well within the ideal range and each covers a distinct part of the governance workflow: insight capture, rule creation, rule listing, outcome reporting, and rule scanning. There is no redundancy.
The surface lacks tools to persist, update, or delete rules—create_rule explicitly writes nothing, and there is no edit or delete operation. While listing and scanning provide read access, the lifecycle is incomplete, forcing manual file operations.