MCP Server Tester

What It Does

The Model Context Protocol (MCP) is the standard for exposing tools to LLMs (Anthropic’s Claude Code, Cursor, Continue, and many other clients use it). MCP servers ship tools whose schemas the LLM reads to decide when and how to call them. Bad schemas = bad tool selection = burned tokens.

This tool catches schema issues before they ship:

1. Paste your server’s tools/list response (or just an array of tools)
2. Validator runs against MCP protocol requirements + LLM-friendliness heuristics
3. See exactly the JSON-context format the LLM would receive
4. Simulate a tools/call to verify arguments validate against your inputSchema

Checks Performed

Spec requirements (errors):

• Missing name field
• Missing inputSchema field
• inputSchema.type must be "object"

Best practices (warnings):

• Name contains chars outside [a-zA-Z0-9_-] (some clients restrict)
• No description (LLM tool selection works better with detailed descriptions)
• Description shorter than 20 chars (under-specified)
• Properties without description (argument purpose unclear)
• Properties without type (LLMs may guess wrong)

Tool-Call Simulation

Pick a tool from the dropdown - the tool auto-generates a starter argument stub from your schema. Edit the JSON, watch the validator catch violations in real time:

• Type mismatches ("5" where "integer" was expected)
• Missing required properties
• Enum violations ("meters" not in ["imperial", "metric"])
• Number range violations (limit: 30 when maximum: 20)
• String length / unknown property warnings

This is the same JSON-Schema check your server probably does in code - run it in the browser before round-tripping through an actual LLM.

What the LLM Sees

The tool also shows you the exact JSON your tools become in the LLM’s context window, formatted as the major clients format it. The token estimate is your per-request cost when the tools are loaded (use Prompt Cache Optimizer to see if you can cache them).

What’s NOT In Scope

• Live server testing - no network calls. Paste-only.
• MCP transport protocols (stdio/SSE/HTTP) - just the schema layer
• OAuth flows - MCP authentication isn’t covered here
• Tool result validation - only input validation

Related Tools

• JSON Schema to LLM Tool Definition - convert plain JSON Schema to MCP/OpenAI tool format
• System Prompt Linter - lint the rest of your agent prompt
• Prompt Cache Optimizer - if your tools are big, cache them
• Reasoning Cost Calculator - what each tool-using request will cost