Guide: URL Resolution and Keyword Search

A key feature of the deepwiki_fetch tool is its ability to accept flexible, user-friendly inputs instead of requiring a precise Deepwiki URL. This guide explains the internal logic that resolves various input formats into a target URL.

This process is primarily handled in src/tools/deepwiki.ts.

The Resolution Flow

The server follows a sequence of steps to interpret the url parameter:

Full URL Check: First, it checks if the input is already a valid HTTP/HTTPS URL. If it is, no further processing is needed.
```
if (/^https?:\/\//.test(url)) {
  // Use URL as is
}
```
Owner/Repo Format: Next, it checks for the common owner/repo format (e.g., vercel/ai). If this pattern matches, it directly prepends the Deepwiki domain.
```
if (/^[^/]+\/[^/]+$/.test(url)) {
  // Already in owner/repo format
  url = `https://deepwiki.com/${url}`;
}
```
Keyword Extraction (NLP): If the input is a free-form phrase (e.g., how can i use vercel ai) or a single word, the server uses Natural Language Processing to find the most likely technology or library name. This logic resides in src/utils/extractKeyword.ts.
- It uses the wink-nlp library to perform Part-of-Speech (POS) tagging.
- It identifies nouns (NOUN) and proper nouns (PROPN) as potential candidates.
- It filters out common stop words like 'how', 'can', 'i', 'to', etc.
- The first plausible candidate is returned.
```
// src/utils/extractKeyword.ts
export function extractKeyword(text: string): string | undefined {
  const doc = nlp.readDoc(text);
  const candidates: string[] = [];
  doc.tokens().each((t) => {
    const pos = t.out(its.pos);
    const value = t.out(its.normal);
    if ((pos === 'NOUN' || pos === 'PROPN') && !stopTerms.has(value)) {
      candidates.push(value);
    }
  });
  return candidates[0];
}
```

GitHub Repository Resolution: Once a keyword is identified, the server queries the GitHub Search API to find the most popular repository matching that keyword. This step, found in src/utils/resolveRepoFetch.ts, translates a name like ai into a canonical owner/repo string like vercel/ai.

// src/utils/resolveRepoFetch.ts
export async function resolveRepo(keyword: string): Promise<string> {
  const url =
    `https://api.github.com/search/repositories?q=${encodeURIComponent(
      `${keyword} in:name`,
    )}&per_page=1`;
  // ... fetch logic ...
  const { items } = (await res.json()) as { items: { full_name: string }[] };
  return items[0].full_name;
}

Final URL Construction: The resolved owner/repo string is then used to construct the final https://deepwiki.com/... URL, which the crawler uses as its starting point.

This multi-step process makes the tool significantly more robust and user-friendly, allowing users to interact with it more naturally.