Returns a per-MCP-server health summary for the agent. Each MCP
server referenced by the agent (its tools.json plus every
subagent’s tools.json) is probed in parallel by issuing a single
MCP initialize JSON-RPC call. Credentials are resolved per
(tenant, server) using the configured authentication method.
Per-server failures surface as
ok=false with an error_code (auth_failed | unreachable |
invalid_response) and a human-readable error string.
mcp_check.ok is true only when every server returns a
healthy response; an empty server list is treated as healthy.
Documentation Index
Fetch the complete documentation index at: https://docs.langchain.com/llms.txt
Use this file to discover all available pages before exploring further.