Production MCP Servers Models Use Correctly
Audit a live MCP server the way a model sees it: descriptions it can route on, schemas it fills right, transport, auth, versioning.
$49
Works with the AI Agents you already use
Production MCP Servers Models Use Correctly
Example session with this skill installed
Audit my MCP server at https://api.internal.dev/mcp and suggest improvements for a tool named 'process_data' that the model keeps confusing with 'analyze_data'.
- Read your context and instructions
- Compiled the production mcp servers
I have completed the audit of your MCP server at https://api.internal.dev/mcp. The valider_mcp.py script identified significant ambiguity in the process_data tool description and missing units in the schema.
MCP Server Audit Report
| Component | Finding | Severity | Recommendation |
|---|---|---|---|
| Tool Description | process_data uses generic text: "Processes a dataset." | High | Apply the WHEN rule to differentiate from analyze_data. |
| Schema Validation | timeout parameter lacks units. | Medium | Rename to timeout_ms or specify units in the property description. |
| Server Instructions | initialize response contains empty instructions. | Medium | Add workflow grammar to prevent the model from processing before validation. |
| Transport | Server currently only supports application/json. | Low | Implement text/event-stream (SSE) for production scalability. |
Refactored Tool Definition
{
"name": "process_data",
"description": "WHEN the user needs to write raw logs to the database after validation is complete. Not for visualization - use analyze_data for charts and summaries.",
"inputSchema": {
"type": "object",
"properties": {
"dataset_id": {
"type": "string",
"description": "The UUID of the validated batch."
},
"retention_days": {
"type": "integer",
"description": "Number of days to store logs (e.g., 30)."
}
},
"required": ["dataset_id"]
}
}
Next steps
- Update the
process_datadescription in your server code to use the "WHEN" trigger provided above. - Add typed error responses to the
process_datafunction to handleidentity_requiredorvalidation_failedstates. - Integrate
scripts/valider_mcp.py --strictinto your CI pipeline to catch regression in tool descriptions.
Connects securely to your tools. The creator never sees your data.
About this skill
An MCP server is a user interface for a model, not an API for you.
Ship an MCP server that models use correctly and that survives real traffic - tool descriptions a model can route on, schemas it fills right, Streamable HTTP transport, authentication and keys, versioning without breaking clients, large files out of the response, typed errors, rate limits and observability. Use when building or hardening an MCP server, when a model keeps calling the wrong tool, or when a server works in a demo and misbehaves in an agent.
The model never reads your documentation. It reads the tool name, the description and the JSON schema, and decides from those three alone. Nearly every "the model is dumb" complaint is a description that says what the tool does instead of when to use it.
Tested scripts: scripts/valider_mcp.py connects to a live server, runs initialize and tools/list, and reports what the model will actually get - missing descriptions, descriptions that never say when, absent or vague schemas, generically named parameters that models fill badly, credentials leaked into a description, too many tools in one server, missing server instructions. It never calls a tool, so it is safe against production.
It was validated against two real servers. One exposes 70 tools where 70 of 70 descriptions open with a trigger word; its sibling exposes 18 where 0 of 18 do - same author, same code quality, opposite routing behaviour. That contrast is the whole argument for the WHEN convention, and the script measures it in seconds.
Includes an EXPERT.md reference (exact parameters, thresholds, failure diagnosis): the description template with the four lines that earn their place, schema naming rules, transport choice, three authentication models and why a key must never travel in a query string, three ways to return large results, errors as a typed interface with a code family table, versioning rules for an API that has no compiler, the observability counters that answer the only two questions you will have, and an eight-row failure table.
Acceptable use: auditing and hardening servers you operate or are authorised to test. The skill refuses probing third-party servers without permission, enumerating tools to find an unprotected one, embedding credentials in descriptions or schemas, and building tools that act irreversibly on a user's behalf without confirmation.
Network: the script makes two read-only JSON-RPC calls - initialize and tools/list - to the URL you give it, and nothing else.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Requires Python 3. Does not execute tools during audits. Designed for standard MCP spec; non-compliant custom extensions may not be fully validated.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 27 days ago
- Software
Python