An MCP server for fetching Wikipedia page content, links, and metadata
WikipediaFetcher is a simple read-only tool suite with three focused utilities. All tools have explicit descriptions and input schemas with types. However, several baseline quality issues reduce the score: (1) Descriptions are borderline, they include Args/Returns boilerplate that duplicates parameter names, adding noise without LLM-relevant context. (2) Parameter descriptions are trivial (e.g., 'The title of the Wikipedia page to fetch' for get_wikipedia_page_text's page_title param) and do not explain constraints, expected format, or error cases. (3) Output schemas are undocumented, the code returns Dict[str, str] and Dict[str, List[str] | str], but the LLM has no visibility into the exact field names ('text', 'error', 'links') or types without inspecting Python code. (4) No error guidance, when a page is not found, the tool returns {"error": "Page 'X' not found"}, but offers no recovery suggestion (e.g., 'try searching with a partial name'). (5) Tools 2 and 3 (get_wikipedia_page_links_titles vs get_wikipedia_page_links_urls) are near-duplicates with nearly identical descriptions, the distinction is only in the response format (titles vs URLs), which is not clear from the names alone. (6) No tool composition or chaining hints, the LLM cannot infer that get_wikipedia_page_links_titles output can feed into get_wikipedia_page_text. (7) No rate-limiting, timeout handling, or pagination despite Wikipedia API's potential for large link lists. (8) No security/audit trail logging, though read-only tools pose minimal risk. Per-tool scores: get_wikipedia_page_text (53), get_wikipedia_page_links_titles (46), get_wikipedia_page_links_urls (45). Average: 48.
Fetches all links from a Wikipedia page using the Wikipedia API. Args: page_title: The title of the Wikipedia page to fetch links from Returns: A dictionary containing the list of linked page titles or error message.
Fetches all links from a Wikipedia page using the Wikipedia API. Args: page_title: The title of the Wikipedia page to fetch links from Returns: A dictionary containing the list of linked page URLs or error message.
Fetches the text content of a Wikipedia page using the Wikipedia API. Args: page_title: The title of the Wikipedia page to fetch Returns: A dictionary containing the page text or error message.
Output schemas undocumented in tool descriptions. LLM cannot see that tools return dicts with 'text'/'links' and 'error' fields without reading Python source code. Rubric requires documented return types (100% of A+ tools). Schema score capped at 0 if not visible in metadata.
Parameter descriptions are trivial (10-15 chars). 'The title of the Wikipedia page to fetch' repeats the parameter name without explaining constraints, format, or error recovery. Rubric baseline: average param annotation is 72 chars; descriptions under 20 chars score 0-20. No mention of maximum length, valid character sets, or what happens if special characters are included.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 37 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
No error recovery guidance. When a page is not found, tool returns {"error": "Page 'X' not found"}, but offers no actionable next step. Rubric requires: 'Error responses must tell the LLM what to do next.' Example: 'Page not found. Try searching with a partial title or check spelling.' Current behavior leaves the agent stuck.
Tools 2 and 3 (get_wikipedia_page_links_titles vs get_wikipedia_page_links_urls) are near-duplicates. Both fetch links from a page; the only difference is the output format (titles vs URLs constructed in the tool). Descriptions are identical. Rubric: 'Avoid multiple tools that do the same thing differently.' Consider a single get_wikipedia_page_links tool with an optional 'format' enum parameter (titles|urls|both) to reduce LLM confusion.
No pagination or result limiting. A Wikipedia page can have hundreds or thousands of links. Tool description does not specify a limit, and the code returns ALL links. Rubric: 'cap results at a reasonable limit (20-50).' Without pagination, a page with 500+ links blows the context window. Add limit and offset parameters, return a total_count, and document the default limit.
No tool composition hints. The LLM cannot infer that links returned by get_wikipedia_page_links_titles can feed into get_wikipedia_page_text to fetch related pages. Descriptions should include dependency guidance: 'Call get_wikipedia_page_links_titles first to discover related pages, then fetch them with get_wikipedia_page_text.' Rubric: 'Include dependency hints.'
Tool descriptions include Args/Returns boilerplate that duplicates parameter names and adds noise. Rubric: 'Write descriptions as if prompt-engineering.' Examples: (1) 'Args: page_title: The title...', this repeats the parameter definition. LLM already sees 'page_title' in the schema; the description should explain WHEN and WHY to use it, not restate it. (2) 'Returns: A dictionary...', generic and unhelpful. Replace with: 'Fetches the full text content of a Wikipedia article. Returns structured data with article text and edit history link. Useful for answering detailed questions or summarizing topics.' Keep descriptions 50-200 chars, focused on intent and context.
No timeout or rate-limiting handling. Wikipedia API calls can hang or be rate-limited. Tool code does not catch timeouts or 429 responses, and descriptions do not mention retry behavior or rate limits. Add explicit timeout (e.g., 10 seconds), handle httpx.TimeoutException, and return a clear error: 'Request timed out. Please retry in a few seconds.' Rubric: 'Set explicit timeouts for tools that call external services.'