MCP Server Tester
Test MCP tool definitions with simulated calls and see what LLMs receive.
- Free
- No account
- Runs in your browser
- Nothing uploaded
Runs entirely in your browser — your input is never uploaded, logged, or stored.Privacy policy
What is MCP Server Tester?
Model Context Protocol (MCP) servers expose tools to LLMs like Claude, GPT, and Gemini. Before shipping a server, you need to verify that your tool definitions work correctly — that the schemas validate arguments, the descriptions guide the model, and the token budget is reasonable.
This tester simulates what happens when an LLM receives your tool definitions: 1. Paste your tool definitions (from tools/list or your code) 2. Select a tool and fill in test arguments 3. See exactly what the LLM would receive in its context window 4. Get a token estimate for the tool definitions
Two modes of testing: • Schema Validation — checks if your tool definitions are valid MCP • Argument Simulation — generates test arguments from your schema and shows what the LLM would see
This is a static tester — it validates your tool definitions and simulates argument generation without connecting to a real MCP server. For testing live server connections, use MCP Playground or similar tools.
This tool performs two types of analysis:
**1. Schema Validation** Validates tool definitions against the MCP specification (revision 2025-11-25): • Checks required fields (name, inputSchema) • Validates inputSchema structure (type must be "object") • Checks property descriptions and types • Estimates token cost (~4 chars per token approximation)
**2. Argument Simulation** When you select a tool and click "Generate Test Arguments": • Reads the tool's inputSchema • Generates realistic test values based on property types and constraints • Shows the exact JSON that would be sent to the LLM as tool definitions • Estimates the token count for the tool definition + arguments
The token estimation uses a ~4 chars per token approximation, which is close to actual tokenizers (cl100k_base for GPT, Claude's tokenizer for Anthropic). The real count varies by tokenizer, but this is close enough for budget planning.
| This tester | MCP Playground | mcpdoctor | |
|---|---|---|---|
| Server connection required | No | Yes | Yes |
| Schema validation | Yes | Yes | Yes |
| Argument simulation | Yes | Yes | No |
| Token cost estimation | Yes | Partial | No |
| Works offline | Yes | No | No |
| No install required | Yes | Yes | No (CLI) |
Worked examples
- Paste a tool definition: {"name":"get_weather","inputSchema":{"type":"object","properties":{"city":{"type":"string","description":"City name"}},"required":["city"]}} → 1 tool found, ~45 tokens
- Click "Generate Test Arguments" → {"city":"San Francisco"}
- The LLM would see: {"tools":[{"name":"get_weather","description":"...","inputSchema":{...}}]}
- Paste a tools/list response with 3 tools → 3 tools found, ~150 tokens total
How to use MCP Server Tester
- Paste your MCP tool definitions — single tool, array, or tools/list response.
- Review the validation results and token estimate.
- Select a tool from the dropdown (if multiple).
- Click "Generate Test Arguments" to see what the LLM would receive.
- Copy the tool definition JSON for use in your MCP server.
Common errors
- Error: "No tool definitions found" — The input must be a tool object, array of tools, or a tools/list response with a "tools" array.
- Error: "Invalid JSON" — Check for syntax errors: missing quotes, trailing commas, or unclosed braces.
- Warning: "Tool has no description" — LLMs use descriptions for tool selection. Add a clear, descriptive sentence.
- Warning: "Property has no type annotation" — Untyped properties force the model to guess the argument type.
FAQ
Does this connect to my MCP server?
No — this is a static tester. You paste your tool definitions, and everything is validated and simulated locally in your browser. No server connection is made. For testing live connections, use MCP Playground.
What's the difference between this and the MCP Schema Validator?
The Schema Validator checks your tool definitions for spec compliance and LLM-friendliness. This tester goes further: it simulates argument generation and shows you exactly what the LLM would receive in its context window.
How accurate is the token estimate?
It uses a ~4 chars per token approximation, which is close to actual tokenizers. The real count varies by tokenizer, but this is close enough for budget planning. For exact counts, use our LLM Token Counter tool.
Can I test a live MCP server?
This tool is for testing tool definitions, not live server connections. For testing live servers, use MCP Playground (mcpplayground.tech) which can connect to your server and run real tool calls.
Related tools
Prefer AllUtil on Google
One click adds AllUtil to your Google preferences. You'll see our tools highlighted with a Preferred badge in Search and AI answers.