Package Registry MCP has 16 well-structured read-only tools covering NPM, Cargo, NuGet, PyPI, Go, and GitHub Security Advisories. Tool naming follows verb_noun pattern consistently (search-*, get-*, etc.), making intent clear. Descriptions are present on all tools and range from 40-100 characters, meeting the minimum bar but remaining brief and generic. Input schemas are visible and properly typed (string, number types with descriptions). However, output schemas are not documented anywhere in the source code, a critical gap for LLMs planning downstream calls. Parameter descriptions are adequate but minimalist. Error handling is absent from visible code (no try-catch, validation, or recovery guidance shown). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. All tools appear to be actually implemented (not inferred), so no cap is applied for that reason.
Tools (16)
get-cargo-package-detailsread only50/100
Gets detailed information about a cargo package.
get-cargo-package-versionsread only50/100
Get version information for a Cargo package
get-github-advisory-detailsread only50/100
Get details of a specific GitHub Security Advisory by its GHSA ID.
get-golang-package-detailsread only50/100
Get detailed information about a specific Go package
get-golang-package-versionsread only50/100
Get version information for a Go package
get-npm-package-detailsread only50/100
Get detailed information about a specific NPM package
get-npm-package-versionsread only50/100
Get version information for an NPM package
get-nuget-package-detailsread only50/100
Get detailed information about a specific NuGet package
Output schemas not documented. No API response structure is documented in code or JSDoc. LLMs cannot plan downstream operations or extract required fields without documentation of what each tool returns.
No error handling or recovery guidance visible in source code. No try-catch blocks, validation logic, or error messages that tell LLMs what to do next (retry, ask user, etc.). This forces agents to guess when failures occur.
Add TypeScript JSDoc comments to each tool handler showing the return type structure. E.g., @returns {Object} { packages: Array<{name, version, downloads}>, total: number, nextCursor?: string }. This documents output schema for LLMs.
Wrap all tool implementations in try-catch blocks with structured error responses. E.g., catch (error) { return { error: 'Package not found. Try searching with a different query.', retryable: true }; }. Include actionable next steps for LLMs.
Add readOnlyHint: true to all tool registrations since all tools are read-only queries. This is a small change (one line per tool in tool registration) that signals safety to LLMs.
Expand tool descriptions to 100-200 characters explaining both WHAT and WHEN. E.g., 'Get detailed metadata for a specific NPM package (name, version, dependencies, license, repository). Use after searching to get full details on a promising result.'
Add min/max constraints to 'limit' parameters in descriptions: 'limit: number (minimum 1, maximum 100, default 20). Results are capped to prevent context window exhaustion.'
Document pagination strategy for search tools. Add a 'page' or 'offset' parameter and document the response structure (total count, next_page_token) so LLMs know how to fetch more results.
Add a README documenting expected output formats for each tool group (npm, cargo, nuget, pypi, go, github-advisories). Include sample JSON responses so LLMs understand field names and types.
Migrate from STDIO to HTTP+SSE or Streamable HTTP transport. STDIO limits the server to local testing only and prevents use with hosted MCP clients. HTTP enables deployment as a cloud service.
Score history
Overall score trend
↑ 18 points across a rubric change (v1 → v2)
53/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
D
53
2026-07-28+
v2
2026-03-09
F
35
-
v1
read only50/100
Get version information for a NuGet package
get-package-advisoriesread only50/100
Get security advisories affecting a specific package.
get-pypi-package-detailsread only50/100
Get detailed information about a specific PyPI package
get-pypi-package-versionsread only50/100
Get version information for a PyPI package
search-cargo-packagesread only50/100
Search for Cargo packages by query string
search-github-advisoriesread only50/100
Search GitHub Security Advisories with optional filters.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All 16 tools are read-only and idempotent, but this is not declared in the tool registration. LLMs cannot automatically determine tool safety without explicit hints.
Descriptions are generic and minimal (40-100 chars). They state WHAT the tool does but lack context for WHEN to use it. E.g., 'Get detailed information about a specific NPM package' does not explain how this differs from search-npm-packages or when an LLM should prefer one over the other.
Parameter 'limit' on search tools has no bounds documented (min/max). Unbounded numeric parameters allow LLMs to pass absurd values (limit: 10000) that break APIs or timeout. Descriptions should state 'limit: number (1-100, default 20)'.
No pagination documented. Search tools may return hundreds of results, but no 'offset', 'page', 'next_cursor' parameters or result count fields are visible. Large result sets risk context window exhaustion.
Add input validation: e.g., check packageName is not empty before making API calls. Return 'Invalid package name: must be non-empty string' rather than letting bad input hit the external API.
Implement rate limiting and timeouts. The request helper has a 10-second timeout (good), but add rate-limit awareness (e.g., respect 429 responses) and return clear guidance: 'Rate limited. Wait 60 seconds before retrying.' instead of raw HTTP errors.