Into the Expanse: A Security Leader’s Guide to the Agent Frontier
Read more
Deploy Claude across your enterprise with continuous environment visibility, pre-inference inspection, and reach containment. Reco governs agent identities, MCP tool permissions, and third-party data reach surrounding Claude across Claude Enterprise, Claude Platform, native inference hooks, and local Model Context Protocol (MCP) connections.
Complete AI governance in both directions: Reco governs the environment surrounding Claude, and Claude queries Reco’s normalized context graph.
Reco connects out-of-band via read-only APIs and native platform hooks to inspect Claude administrative, execution, and development surfaces (supported for Anthropic-hosted deployments; does not apply to third-party hosted cloud instances such as AWS Bedrock or Google Vertex AI):
Continuous visibility into user directories, enterprise workspaces, project data boundaries, conversation content inspection, and user activity feeds. Enforces centralized off-boarding, MFA policies, and prevents sensitive data exfiltration.
Moves security enforcement directly onto Anthropic infrastructure. Reco acts as your configured security evaluation server: prompts, prior tool calls, and session transcripts are evaluated against corporate policy and cleared or blocked before Claude ever sees them—stopping PII, payment card numbers, or proprietary code leaks across Claude Chat, Cowork, and Claude Code sessions.
Complete governance over developer API keys, workspace memberships, and activity/audit logs. Tracks assigned tools, service accounts, and permission policies back to human owners without citing or restricting underlying model execution.
Inspects and governs Model Context Protocol (MCP) client-server connections across local developer endpoints (`stdio`) and remote services (`SSE`). Prevents tool poisoning, tool drift, and unauthorized tool invocation surrounding Claude.
Reco exposes a native Model Context Protocol (MCP) server that enables Claude to query enterprise security posture:
Unsanctioned Claude accounts, rogue projects, and unvetted MCP servers deployed by business units.
Continuous discovery across endpoints, identity providers, and network egress; classifies every Claude workspace automatically.
Unregulated transfer of customer records, PII, intellectual property, or source code into Claude prompts.
Real-time data exposure mapping via Claude Enterprise and pre-inference blocking via native inference hooks before data reaches the model.
Malicious instructions embedded in untrusted external data (memory/RAG poisoning or compromised MCP responses).
Runtime MCP tool-call inspection, schema validation, and third-party data reach boundary enforcement around Claude.
Long-lived Claude Platform API keys hardcoded in codebases, CI/CD pipelines, or local scripts.
Instant detection of leaked, dormant, or over-scoped API keys with automated revocation workflows.
Autonomous agents granted broad read/write access beyond their immediate task scope.
Enforces least-privilege boundaries and maps transitive blast radius across all downstream third-party API connections surrounding Claude.
Departing engineers or contractors leaving behind active Claude Platform API keys, projects, or autonomous agents.
Correlates IdP lifecycle status (Okta, Azure AD, Ping) to immediately revoke orphaned tokens and reassign agent ownership.
While point AI security tools treat Claude as an isolated silo, Reco analyzes the broader environment:
CC6.1, CC6.2, CC6.3 (Access control, least privilege, off-boarding)
Verifiable inventory of all Claude users, active Claude Platform API keys, inference policy logs, and automated de-provisioning records.
A.9.2, A.9.4 (User access management and system access restriction)
Continuous configuration drift detection and RBAC enforcement across Claude Enterprise.
GOVERN 1.2, MAP 1.5, MEASURE 2.6 (AI risk mapping and reach containment)
Multi-hop blast-radius mapping and continuous non-human identity lifecycle monitoring across the Claude environment.
Control 5 (Account Management) & Control 6 (Access Control Management)
Full discovery of shadow Claude accounts, unapproved workspaces, and stale API tokens.
