MCP server for interacting with Dev.to API to manage articles and retrieve content
This server has 5 tools with mixed quality. Three read-only tools (GetLatestArticles, GetTopArticles, GetArticlesByTag) have minimal schema, empty input objects with no parameters. Two tools (GetArticleById, CreateArticle) have properly typed parameters with descriptions. However, critical issues undermine the overall quality: (1) Tool descriptions are vague and lack guidance on when to use each tool vs alternatives. GetLatestArticles and GetTopArticles both retrieve articles but descriptions don't clarify the distinction. (2) Output schemas are completely undocumented, we cannot see what fields the tools return or what structure the agent should expect. The service returns string representations (FormatHelper.FormatArticles) rather than structured objects. (3) No error handling documentation, no guidance on recovery, retries, or what happens on API failures. (4) The CreateArticle tool lacks confirmation/dry-run pattern for this destructive operation. (5) Parameters like 'tags' use comma-separated strings instead of arrays, which is non-idiomatic and harder for LLMs to compose.
Create a new article on Dev.to
Get article details by ID
Get articles by tag from Dev.to
Get latest articles from Dev.to
Get top articles from Dev.to
Empty input schemas for three read-only tools (GetLatestArticles, GetTopArticles, GetArticlesByTag initially, though GetArticlesByTag does accept a 'tag' parameter). The first two tools claim to have {} inputs but no parameters are exposed, making them indistinguishable in intent to an LLM.
Output schemas completely undocumented. All tools return Task<string>, formatted text blobs. The agent cannot know what fields to extract, what structure to expect, or how to chain results to downstream tools. Returning unstructured text wastes tokens and invites parsing errors.
No error handling guidance. CreateArticle and GetArticleById call external APIs (dev.to) with no documented recovery strategy. What happens on 401 (auth failure), 404 (article not found), or timeout? LLM gets no direction.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 10 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 27 | - | v1 |
Ambiguous tool naming and descriptions. GetLatestArticles vs GetTopArticles are not clearly distinguished. Does 'latest' mean chronologically newest? Does 'top' mean trending/popular by engagement? Description does not clarify, forcing LLM to guess.
CreateArticle lacks destructive operation confirmation pattern. Agents can publish articles to a live blog without any confirmation or dry-run option. This is high-risk for accidental content publication.
Parameter 'tags' in CreateArticle uses comma-separated string format instead of an array. This is non-idiomatic and harder for LLMs to construct correctly. Example: tags='javascript,webdev' vs tags=['javascript', 'webdev'].
No pagination support documented. GetLatestArticles and GetTopArticles return exactly 10 items (per .Take(10) in code), but no offset/page/limit parameters exposed to the LLM. If the agent needs more articles, it has no way to request them.
API key hardcoded in constructor via IConfiguration. While better than inline strings, no documentation on how the key is injected or what permission scopes it grants. If leaked, the scope of compromise is unclear.