Engineering workflows
Claude Code
Assess proposed repository deletions, permission changes, production deployments and other engineering actions before your tool runner executes them.
Flagship use case
AI Agent Action Security for Claude
Claude Code. Claude Cowork. claude.ai. One Asyrx intelligence layer for the proposed action — before it happens.
Connect your approved Enterprise Sandbox through account-based MCP access. Asyrx returns an Action Risk Score, a Blast Radius Score, Reason Codes, Behavioral Context, Policy Context and Recommended Controls. Your enterprise retains final authorization and execution control.
Knowledge-work workflows
Claude Cowork
Evaluate the scope of proposed file operations, sensitive-data movement and cross-system changes before the connected workflow carries them out.
Conversational workflows
claude.ai
Bring proposed actions into an Asyrx sandbox assessment from a conversation and read saved test assessments with your own workspace permissions.
Before execution
Claude proposes. Asyrx assesses. Your enterprise decides.
A connected assistant can ask for an assessment. Enforcing that assessment before every consequential action is the responsibility of your orchestrator or tool runner — connecting MCP alone does not enforce a gate.
- 01
Describe the proposed action, target system, environment and expected scope.
- 02
Call assess_action for a test-only risk preview using the same engine as the Asyrx API.
- 03
Apply your own policy and human-review requirements before any external action executes.
Protected connection
Connect your approved sandbox
Use the same remote MCP address in all three Claude experiences. Sign in to Asyrx and review the account connection request; no Asyrx API key needs to be pasted into a conversation.
Remote MCP server URL
https://www.asyrx.com/mcpClaude Code
Terminal
claude mcp add --transport http asyrx https://www.asyrx.com/mcpOpen Claude Code, use /mcp to check the connection and complete the browser sign-in when prompted.
Claude Cowork and claude.ai
In Claude, open Customize → Connectors → Add custom connector and enter the server URL. On Team or Enterprise, an organization Owner adds the custom connector in Organization settings → Connectors; members then connect their own Asyrx accounts. Availability and organization settings are controlled by Anthropic.
Asyrx must be republished with this integration before the new connection can be used on the public domain.
Sandbox tools
Risk intelligence, not execution access
- list_sandboxes
- Find your approved, active enterprise sandboxes. No anonymous access or automatic workspace provisioning.
- assess_action
- Preview a proposed action using test-only behavioral history and your workspace policy context. Previews are not saved and do not trigger webhooks.
- list_assessments
- Read recent saved test assessments in your workspace. Live and demonstration records are excluded.
For persistent assessments, request IDs, idempotency and API-key-based server-to-server workflows, use the existing POST /api/v1/actions/assess endpoint with an approved test key. MCP previews do not replace that API contract.
Across enterprise systems
One assistant, many consequential actions
Claude is the agent interface; Salesforce, Shopify and GitHub are examples of target systems. Assess the actual target and scope rather than treating the assistant's name as a risk signal.
This is a custom Asyrx integration, not an Anthropic partnership, endorsement or directory listing. Asyrx does not send your assessment metadata to an AI model; Claude's handling of your conversation and tool results is governed by your Anthropic account and policies. Never submit secrets or unnecessary sensitive content.