MCP API
EvidentSource speaks the Model Context Protocol. AI agents configured with the MCP endpoint can discover and use EvidentSource’s capabilities as native tools.
See MCP for AI agents for the Sandbox setup.
Endpoint
Section titled “Endpoint”- Sandbox:
https://sandbox.evidentsource.com/mcp - Your deployment: typically at
/mcpunder your ALB, behind whatever authentication you’ve configured
The MCP server exposes:
| Tool | Purpose |
|---|---|
list_databases | Enumerate databases accessible to the caller |
list_commands | Enumerate commands in a database |
issue_command | Send a command |
list_state_views | Enumerate state views in a database |
read_state_view | Read a specific state view with optional bi-temporal params |
query_events | Query event streams with filters |
list_components | Introspect State Change / State View components |
Authentication in production
Section titled “Authentication in production”The Sandbox’s MCP endpoint is open. Your production endpoint sits behind your configured identity provider — tokens are exchanged for MCP session credentials via token vending.
Why this is interesting
Section titled “Why this is interesting”Because every command and every event the agent produces carries a full decision trace, an agent acting through MCP is auditable by construction. You don’t need to instrument the agent or its prompts — the system captures what happened and why.