Skip to main content
The MCP Explorer is a top-level console route at /mcp (promoted from settings in Phase 7). It surfaces the platform’s Model Context Protocol server so you can see exactly what your LLM clients are exposed to.

​
Layout

A header strip with live counts (“42 Tools / 7 Resources / 3 Prompts”) followed by a three-tab body:
TabContents
ToolsEvery MCP tool the platform exposes — name, description, and the first few input-schema fields
ResourcesURI-addressable resources (run bundles, ledger exports, intervention payloads)
PromptsServer-managed prompt templates that LLM clients can fetch
Click any row for the full schema in a side panel.

​
Tools tab

Each tool card shows:
  • Name (mcp_tool_name) — what the LLM calls
  • Description — server-supplied summary
  • Input schema — first three required fields, with overflow indicator (+5 more)
Selecting a tool opens a detail view with the complete JSON Schema and example invocations.

​
Resources tab

Resources are URI-addressed read-only payloads — typically run bundles (runs/{run_id}), ledger snapshots, and intervention details. Useful for inspecting what an LLM client can pull without executing a tool.

​
Prompts tab

Server-managed prompts the LLM can request by name. Authoring lives in the platform; this tab is read-only.

​
Why a top-level route?

Phase 7 promoted MCP from a settings sub-page to a top-level entry to match how teams actually use it: MCP discovery is part of debugging an agent, not a configuration task. The new sidebar lists /mcp next to /runs and /ledger so it’s reachable in one click during incident triage.

​
Underlying API

The page reads from the MCP HTTP gateway:
  • GET /mcp/tools
  • GET /mcp/resources
  • GET /mcp/prompts
Full reference: Platform → MCP.

​
See also