MCP Server Probe

Point it at any Model Context Protocol endpoint and get the full picture in one call: does it speak MCP, what server is it, which protocol version, what capabilities does it declare, how fast does it initialize, does it manage sessions, does it need auth, and — the part you actually came for — every tool it exposes, with each tool's description, input properties, and required fields.

Why this exists

The MCP ecosystem is growing faster than anyone can evaluate it. Before an agent adopts a server (or before you list a dependency on one), someone has to answer: is this endpoint real, healthy, and what exactly can it do? Doing that by hand means writing the initialize/initialized/tools-list handshake yourself. This Actor is that handshake, productized.

We built it because we run our own public MCP server and needed to verify it from the outside, the way a stranger's agent would see it.

What you get

A summary row:

{
  "ok": true,
  "url": "https://gozerai.com/mcp/library",
  "server_name": "1450 Library",
  "protocol_version": "2025-06-18",
  "initialize_latency_ms": 456,
  "session_managed": true,
  "capabilities": ["prompts", "resources", "tools"],
  "tools_count": 3
}

Plus one row per tool:

{
  "name": "search_library",
  "description": "Search a curated library of practical small-business guides...",
  "input_properties": ["query", "limit"],
  "required": ["query"]
}

Input

Field	Type	Description
url	string	The MCP streamable HTTP endpoint
bearerToken	string, optional	Sent as Authorization: Bearer for authenticated servers
timeoutSecs	integer	Per-request timeout, default 30

Behavior and honesty

• Speaks the streamable HTTP transport (JSON and SSE responses both handled), protocol 2025-06-18.
• A non-MCP endpoint, an auth wall you did not provide a token for, or a dead server produces a clean ok: false row with the reason, and you are not charged. Billing is per completed probe.
• Read-only by construction: it initializes, lists, and leaves. It never calls the server's tools.

Use cases

• Agent builders vetting third-party MCP servers before wiring them in.
• Server authors monitoring their own endpoint from outside (schedule it; alert if ok flips or latency spikes).
• Directories and researchers inventorying what a set of MCP endpoints actually expose.

Built and used daily by 1450 Enterprises — this probe's first production target was our own public library server.