OPERATIONAL SAFETY FOR AGENTIC AI

Know what your AI connects to.
Control what it can do.

Raksha Scan discovers and AGP governs.

LOCAL DISCOVERY

Raksha Scan NEW

Your AI connections.
In one local report.

Discover configured MCP servers, trace them to their clients, and review configuration issues.

Explore Raksha Scan macOS · Open source · Offline
RUNTIME GOVERNANCE

AGP

Your agents act.
You set the rules.

Govern MCP actions with policies, human approvals, credential custody, and an attributable audit trail.

Explore AGP Self-hosted · Your infrastructure

Use either product independently, or bring discovery and governance together.

RAKSHA SCAN · DISCOVER

See the connections
behind your AI clients.

MCP configurations spread across tools and projects. Bring them into one local inventory you can explore.

Illustrative map: AI client configurations become one MCP inventory with AGP, Direct, and Unknown routing indicators.
Illustrative configuration map · AGP routing is inferred from configuration.
DISCOVER

Map clients to servers.

Find supported global configurations and the project roots you select. Bring multiple AI clients into one inventory.

UNDERSTAND

Trace every declaration.

Inspect launch details, sanitized endpoints, and the source files behind each configured connection.

REVIEW

Know what needs attention.

Review floating package versions, credential-like key names, plaintext HTTP, and incomplete configurations.

Take a closer look at Raksha Scan No account. No telemetry. No server execution.

AGP · Operational Safety and Governance

Control your agents’ actions.
Before they reach your systems.

Reads can flow. Sensitive actions can wait for a human.
Every decision has an identity and a record.

HELD FOR APPROVALRECORDED SCENARIO

A trade is waiting for a decision.

Agent
claude-desktop-agent
Action
robinhood__place_equity_order
Order
BUY 5 × MRVL · $1,230.35
Policy
Trades require human approval

Interactive replay of a documented AGP session.

THREE CONTROLS. THREE RECEIPTS.

01

The everyday gate

Hold an action for explicit approval before execution.

POLICY_HOLD · HTTP 423
02

The emergency brake

Suspend an agent’s behavior profile during a session.

BEHAVIOR_PROFILE_DENIED
03

The organization-wide kill switch

Remove a tool server from the governed registry.

MCP -32602: unknown tool
Tool governanceBehavior profilesPolicy decisionsHuman approvalsCredential custodyAudit attribution
Naveen Vandanapu, founder of Raksha AI

THE FOUNDER

Built by
Naveen Vandanapu.

25 years building platforms that survive production. Named inventor on 14 patents, with two new provisional patent applications filed at Raksha AI.

“You don’t hope a system behaves.
You engineer it to.”

Raksha brings that principle to agent infrastructure: understand the connections, then put control in the path of action.

YOUR AI. YOUR CONTROL.

Understand your connections.
Take control of the action.