Hosted Enterprise MCPAgentKeeper MCP Server
Give AI assistants a governed way to retrieve AgentKeeper security evidence: redacted by default, policy-aware, bounded, linked back to the console, and audited on every request.
Assistant answer
In the last 24 hours AgentKeeper saw 18 activity events, 2 warnings, 1 blocked MCP tool call, and 1 unmanaged host with gateway drift. Raw prompts and tool I/O were redacted under the standard profile. Open the linked event before declaring impact.
Why it wins
Built for the moment after everyone connects AI to production evidence.
Most MCP servers expose objects. AgentKeeper exposes usable security context with scope boundaries, redaction proof, omissions, and audit trails that security teams can defend.
SOC copilots
Ask for bounded activity, blocked actions, policy outcomes, actor hints, and console links without handing the assistant raw prompts.
GRC evidence
Prepare audit-ready packets with time windows, omissions, redaction counts, result IDs, and durable evidence links.
MCP drift review
Compare runtime-only, unmanaged, deprecated, and unapproved MCP servers across affected hosts.
SOAR workflows
Start with read-safe context, then graduate to governed acknowledge, assign, comment, exception, and export requests.
Redaction as product
Every response reports what was redacted, omitted, and capped. Standard mode never emits raw prompts, tool I/O, secrets, webhooks, or payload bodies.
Setup that feels current
Hosted URL, metadata endpoint, visible-once token, scoped service accounts, compatibility snippets, live test, and audit replay in Settings.
Evidence that assistants can use
Shared response envelopes carry result IDs, time windows, links, summaries, evidence arrays, omissions, and next investigative questions.
Read beta first. Governed writes next.
Start with evidence retrieval, investigation briefs, posture, drift, and policy explanation. Governed acknowledgement, comments, exception requests, and export jobs sit behind explicit scopes and audit.