External-agent access · PlannedYour agent.
Your agent.
Seer underneath.
A shared tool layer for agents that run in your own environment. Read the interface design and the permissions behind it.
Read the architecture| Interface | Purpose | Design |
|---|---|---|
| Typed tools | Discover capabilities and request structured evidence. | Versioned inputs, outputs and source timestamps. |
| HTTP API & MCP | Use Seer from a compatible external agent. | Workspace-scoped authentication and execution-time grants. |
| Persistent events | Keep a watch running while your client is offline. | Cursor-based recovery and signed webhook delivery. |
| Action proposals | Prepare an action for review. | Account, instrument, exact size, expiry and authorization reference. |
One permission model.
Hosted and external agents should use the same domain services. An external client cannot expand access by declaring that an action was approved.
Build against verified contracts.
The local workspace already enforces research-tool grants. Public API and MCP onboarding are planned. This page describes the target interface, not a deployed endpoint or SDK.