An observable, governable, and recoverable Agent runtime platform with tool policy management and execution governance
Synapse is a gRPC-based agent platform, not an MCP server. The 5 tools (GetToolPolicy, PutToolPolicy, ReloadToolPolicy, ListAdminTools, ListToolCatalog) are internal gateway handlers for policy management. Tool definitions are inferred from handler names and brief descriptions in source comments, not from explicit MCP tool registration. No JSON Schema input definitions are visible in the provided code. Descriptions are present but minimal (10-50 chars), lacking WHEN/WHY context. No output schemas documented. This is a policy control plane, not a chat-facing agent tool interface.
Retrieves the current tool policy configuration, either from managed storage or runtime default
Lists all available tools with full administrative metadata including risk levels, approval requirements, and role permissions
Lists available tools filtered by user role and current policy, showing only tools the authenticated user is allowed to use
Updates and applies tool policy configuration, persisting changes and applying them to the runtime
Reloads and reapplies the managed tool policy to the runtime
No explicit MCP tool registration visible. Tool definitions are inferred from handler names and comments, not from formal schema declarations. Cannot verify input/output schemas.
Input schema for PutToolPolicy visible in description but not as formal JSON Schema with type constraints. role_allow, approval_required, disabled_tools lack type validation rules, min/max, or enum constraints.
Descriptions are 10-50 characters, below the 34-char baseline minimum. No WHEN/WHY context. E.g., 'Reloads and reapplies the managed tool policy to the runtime' lacks guidance on when an agent should call this vs GetToolPolicy.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 26 | 2026-07-28+ | v2 |
No output schemas documented. ListAdminTools and ListToolCatalog return tool lists but structure is not specified. Agents cannot plan downstream calls without knowing response fields.
No error handling guidance. PutToolPolicy and ReloadToolPolicy are write operations but no recovery hints provided (e.g., 'If policy validation fails, check role names against ListAdminTools').