Agent Builder
The Agent Builder section in the admin sidebar is a read-only registry view of the agents that power the Anter platform itself. It shows the topology, configurations, and tool schemas of the system's code-based agents — the orchestrators and specialists that handle things like meta-agent builds, skill authoring, and platform operations.
This is not the Flows builder
The Agent Builder admin view is different from Flows. Flows is where you build, configure, and run your own agents. The Agent Builder admin view is a live inspection window into the platform's own internal agent architecture — you can read it but not edit it here.
Three views
The Agent Builder page has three tabs:
| Tab | What it shows |
|---|---|
| Topology | An interactive graph showing how the platform's agents connect to each other and to MCP infrastructure |
| Registry | A card-per-agent grid with full configuration details for each agent in the registry |
| MCP Tools | A reference listing of all tools exposed via MCP, with input schemas |
A status badge in the page header indicates whether the data is Live (served directly from the running registry), Fallback (a cached copy), or Static (compiled configuration).
Topology
The topology graph arranges agents into four rows:
- Standalone agents — agents that operate independently, not orchestrated by another
- Orchestrator — the primary coordination agent that routes to specialists
- Sub-agents & services — specialist agents and domain services the orchestrator delegates to
- MCP infrastructure — the MCP clients that provide tool access to the agents above them
Node types are color-coded:
| Color | Node type |
|---|---|
| Purple | Orchestrator |
| Blue | Sub-agent |
| Gray | Standalone agent or MCP client |
| Green | Domain service |
Edges (the connecting lines) show relationships:
| Style | Meaning |
|---|---|
| Solid purple | Tool delegation — read operation |
| Solid orange | Tool delegation — write operation |
| Dashed gray | MCP client connection |
Click any node to open a detail panel on the right. The detail panel shows the agent's type, version, model, capabilities, endpoints, tools it exposes, and the MCP clients it uses.
Registry
The Registry tab shows the same agents as a grid of cards — one per agent. Each card includes:
- Type badge (Orchestrator / Sub-agent / Standalone)
- Version number
- Name and description
- Model and temperature
- Capabilities
- Tools the agent exposes (purple for read operations, orange for write)
- The API endpoints it is accessible through
- Its architectural pattern and source file location
MCP Tools
The MCP Tools tab lists every tool available across all MCP clients in the registry, split into Read tools and Write tools.
Each tool entry shows:
- Tool name
- Whether it performs a read or write operation
- A plain-language description
- Which service or agent uses it
- An expandable Input schema table — property names, types, which are required, and a description of each
This tab is useful when diagnosing tool-related issues or understanding exactly what parameters a tool expects.
Viewing system prompts
For agents that expose their prompts, a View Prompt button appears on the agent card and in the topology detail panel. Clicking it opens a modal with the full system prompt, a character count, and a Copy prompt button.