Documentation

Connections#

A connection represents a link between an AI agent (or human using an AI tool) and an upstream service. Connections are the primary unit of governance in FirstOps.

Connection Lifecycle#




  
    
  
  
    
  






  
  Daemon scan


  
  Reported to FirstOps


  
  Fix applied


  
  Config reverted


  


  
  Discovered


  
  Unmonitored


  
  Monitored

Discovery#

The FirstOps daemon continuously scans the local machine for MCP server configurations across all supported clients:

  • Claude Code
  • Claude Desktop
  • Cursor
  • VS Code
  • Cline
  • Windsurf
  • Zed
  • JetBrains IDEs

Unmonitored#

A discovered connection that routes directly to the upstream service. Requests bypass FirstOps — no policy enforcement, no credential brokering, no audit trail.

Monitored#

A connection that routes through the FirstOps MCP gateway. Every request is:

  1. Authenticated (DPoP verification)
  2. Authorized (policy evaluation)
  3. Credential-brokered (raw tokens injected at proxy time)
  4. Audited (full request/response logged)

Fixing a Connection#

When you "fix" an unmonitored connection, the daemon:

  1. Requests safe connection parameters from FirstOps
  2. Rewrites the local MCP configuration file
  3. The connection now routes through the FirstOps gateway
# Fix a specific connection
fo discover fix <connection-id>

# Fix all unmonitored connections
fo discover fix --all

Credential Brokering#

The gateway enforces a strict boundary: agents never see upstream credentials.

When a connection is monitored:

  • OAuth tokens and API keys are stored in FirstOps
  • The proxy injects authorization headers at request time
  • The agent's request contains only its DPoP proof
  • Upstream credentials are scoped and rotated independently

This means revoking an agent's access is instant — revoke the principal, and no credential cleanup is needed on upstream services.

Outbound Sets#

An Outbound Set controls which upstream tools the members of an Access Group can reach. You configure it in the dashboard under your Access Group settings:

  • Each Access Group has one Outbound Set
  • The Outbound Set lists permitted upstream services
  • Requests to services outside the Outbound Set are blocked
  • This is evaluated before policy rules