Testnizer
EN TR
Download

MCP (Model Context Protocol)

Testnizer is an MCP client. Connect to MCP servers over Streamable HTTP, SSE, or stdio; list and invoke tools.

The Model Context Protocol is the open spec — published by Anthropic and adopted by Claude Desktop, IDE extensions, and other AI hosts — that lets a language-model host talk to external tools in a standard way. An MCP server exposes a list of tools (with JSON schemas for their inputs); an MCP client (Testnizer, in this case) discovers them and invokes them.

Testnizer is an MCP client. It connects to MCP servers, lists their tools, and lets you invoke any tool with arguments. Use it to:

  • Verify that a new MCP server you’re building responds correctly to the protocol handshake and tools/list.
  • Try out a third-party MCP server before wiring it into Claude Desktop.
  • Smoke-test tool invocations during development without spinning up a host.

Testnizer is built on the official @modelcontextprotocol/sdk, so the wire protocol is bit-for-bit identical to what other hosts speak.

Opening an MCP tab

Click + New → MCP. The editor opens with a transport selector, a URL or command field, a Connect button, and an empty tools list.

Choosing a transport

MCP defines three transports. Testnizer supports all of them.

Streamable HTTP

The current MCP HTTP transport. Use this for remote MCP servers exposed at a URL like https://mcp.example.com/mcp.

  • Transport: Streamable HTTP
  • URL: the full server URL (path included)
  • Connect kicks off the handshake — Testnizer is identified as Testnizer / 1.0.0 to the server.

SSE (legacy)

The earlier HTTP+SSE-based transport, kept for compatibility with older servers. Use this if the server documentation explicitly mentions SSE.

  • Transport: SSE (legacy)
  • URL: the SSE endpoint URL

stdio (local subprocess)

For MCP servers distributed as command-line tools (e.g. on npm). Testnizer spawns the server as a child process and speaks JSON-RPC over its stdin/stdout — no network involved.

  • Transport: stdio (local)
  • URL field: the full command line, e.g. npx @modelcontextprotocol/server-everything or node /path/to/my-mcp-server.js

The first whitespace-separated token is treated as the executable; the rest are passed as arguments. Environment variables are inherited from Testnizer’s process environment.

Connecting

Click Connect. Testnizer:

  1. Opens the chosen transport (HTTP request, SSE stream, or subprocess).
  2. Performs the MCP initialize handshake.
  3. On success, the server’s reported name and version (getServerVersion()) appear next to the Connect button, which becomes a red Disconnect.
  4. The tools list on the left auto-populates from tools/list.

Tools are auto-listed immediately after a successful connection — you don’t need to click a separate “List tools” button.

If the handshake fails, the error is shown beside the connection bar (unreachable URL, command not found, version mismatch, etc.).

The tools list

Every tool the server advertises appears in the left panel:

  • Name — the tool identifier the model would reference
  • Description — the human-readable description from the server, if provided
  • Input schema — when you select a tool, its JSON Schema is rendered above the arguments textarea so you can see what the server expects

Invoking a tool

  1. Click a tool to select it.
  2. Edit the Arguments (JSON) textarea. When you select a tool that has an input schema, Testnizer automatically pre-fills the arguments textarea with a skeleton JSON generated from the schema — strings become "", numbers become 0, booleans become false, arrays become [], and enums get the first value. Edit the values you need; the rest can stay at their zero values. Variables resolve here too ({"path": "{{workspaceRoot}}/file.txt"}).
  3. Click Invoke <tool-name>.
  4. The structured result from the server appears in the Result pane below. MCP results are returned as a typed payload (text, JSON, image references, etc.), all pretty-printed.
  5. If the server returns an error (validation failure, runtime error, etc.), the error message is shown in red instead of the result.

Disconnecting

Click Disconnect to close the transport. For stdio transports this also terminates the subprocess. The tools list is cleared; the last result is preserved on screen until the next invocation.

Multi-tab isolation

Each MCP tab maintains its own connection, transport choice, tools list, selected tool, and last result. Open multiple tabs to compare two MCP server implementations side by side, or to keep a long-running stdio server connected in one tab while iterating against an HTTP server in another.

Security notes

  • stdio transport spawns processes. Only invoke commands you trust — Testnizer cannot sandbox the child process beyond what the host OS provides.
  • Remote MCP servers see Testnizer’s IP address and any arguments you pass to a tool. Treat the same as any HTTP API call to an external service.
  • No secrets are sent automatically. Authentication for HTTP MCP servers must be wired into the URL itself (e.g. with a token in a query parameter) if your server requires it; native MCP auth flows are evolving and will be added as the spec stabilises.