Improve Your Score

Quick Wins

These changes typically take under an hour and can significantly boost your score:

1. Add descriptions to every tool parameter. Schema Interpretability evaluates whether AI models can understand your tool definitions. Missing descriptions are the #1 cause of low scores.
2. Add a README with usage examples. Documentation & Maintenance scoring rewards clear README files with installation instructions and usage examples.
3. Return structured errors, not empty responses. Protocol conformance checks whether your server returns proper MCP error codes instead of silent failures.

Schema Quality

Weight: 20–25% — Are your tool definitions valid and complete?

• Use inputSchema with full JSON Schema for every tool (type, description, required fields).
• Provide a clear, actionable description on the tool itself (not just the parameters).
• Avoid vague parameter names like data or input — be specific.
• Validate your schema with the MCP Scoreboard CLI before publishing.

Protocol Conformance

Weight: 15–20% — Does your server follow the MCP specification correctly?

• Respond to initialize, tools/list, and ping correctly.
• Return proper JSON-RPC error objects for invalid requests (not HTTP errors or empty responses).
• Handle unknown methods gracefully with MethodNotFound errors.
• For remote servers: respond within 10 seconds. Timeouts tank your protocol score.

Reliability

Weight: 5–10% — Is your remote server consistently reachable?

• Only applies to remote (HTTP/SSE) servers. Local stdio servers are N/A.
• Uptime is measured via periodic probes. Ensure your server stays running.
• Use a process manager (systemd, Docker restart policy, cloud auto-scaling) to recover from crashes.
• Consider a CDN or load balancer for high-traffic servers.

Documentation & Maintenance

Weight: 15–25% — Is your project well-maintained and documented?

• Write a README with: what the server does, how to install it, how to use it, and configuration options.
• Include a LICENSE file (MIT, Apache 2.0, etc.).
• Keep dependencies up to date. Outdated dependencies lower your maintenance score.
• Commit regularly. Repos with no activity for 6+ months score lower.
• Tag releases with semantic versioning.

Security Hygiene

Weight: 20–33% — Does your code follow security best practices?

• Never hardcode secrets, API keys, or credentials in source code.
• Validate and sanitize all user inputs before passing to system calls.
• Use parameterized queries for database access (no string concatenation).
• Follow the principle of least privilege — request only the permissions you need.
• Note: this measures implementation patterns, not runtime safety. See our security caveat.

Schema Interpretability

Weight: 10% — Can AI models understand and use your tools correctly?

• Three different AI models independently evaluate your tool schemas.
• Write descriptions as if explaining to someone who has never seen your API before.
• Use consistent naming conventions across tools (e.g., all snake_case or all camelCase).
• Include enum values where applicable instead of free-text string parameters.
• Provide default values for optional parameters.

Claim Your Server

After making improvements, claim your server on its detail page to unlock owner tools including priority rescoring. Claimed owners can request a rescore every 24 hours to see updated results after pushing improvements.

All servers are also automatically rescored on a rolling basis as part of our score lifecycle.

Test Before You Ship

Use our Pre-flight Check to test your remote endpoint instantly — no account required. The CLI tool also provides local schema validation for development.