Model Context Protocol server implementation for Dev.to to allow AI assistants to access and interact with Dev.to content
This server demonstrates weak definition quality across all dimensions. Tool names are generally good (verb_noun pattern), but descriptions are superficial and lack context for LLM decision-making. Input schemas are present but sparse, most tools have minimal parameter documentation. Output schemas are completely undocumented. Error handling is minimal. The codebase shows basic structure but lacks production-grade rigor. 9 of 10 tools have descriptions under 100 characters, well below the 194-char baseline for A+ tools. No tools document their return structure or error cases.
Create and publish a new article on Dev.to
Get a specific article by ID from Dev.to
Get detailed information about a specific article
Get articles by tag from Dev.to
Get articles written by a specific user
Get the latest articles from Dev.to
Get the top articles from Dev.to
Output schemas completely undocumented. No tools document what fields they return or how to parse responses. Tools return formatted strings rather than structured JSON, forcing LLMs to parse unstructured text.
Descriptions are generic and lack context. 'Get the latest articles from Dev.to' does not explain when to use this vs get_top_articles, what structure is returned, or whether results are paginated. Descriptions should be 50 - 200 chars and answer: what, when, and what to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 41 | - | v1 |
Get information about a Dev.to user
Search for articles on Dev.to
Update an existing article on Dev.to
No error recovery guidance. Tools call raise_for_status() and will fail with HTTP errors. Responses do not guide the LLM on recovery: is the error retryable? Should the user be asked? This violates the recovery-guide pattern.
API key exposure risk in create_article and update_article. Both fetch os.getenv('DEV_TO_API_KEY') at runtime. If this server is used in a multi-tenant environment or hosted setting, the key could be leaked in error messages or logs.
No pagination metadata returned. format_articles() limits to 10 items hard-coded but does not return a total count, cursor, or 'more available' indicator. Agents cannot know if results are truncated and cannot request the next page.
Responses are formatted as markdown strings rather than structured JSON. format_articles(), format_article_details(), and format_user_profile() all return plain text. LLMs must parse this text to extract fields like article_id, which wastes tokens and introduces errors.
Destructive operations (create_article, update_article) lack pre-execution confirmation or dry-run capability. An agent could irreversibly create or modify content without warning.
Parameter 'id' in get_article_by_id is typed as string but should likely be integer to match article_id elsewhere. Inconsistent typing (id vs article_id) across tools creates ambiguity for LLMs.