Slice.js Documentation MCP

Instructions

Implementation Reference

• src/tools/get-llm-full-context.ts:4-53 (handler)

  Main implementation of get_llm_full_context tool. Defines the tool with its name, description, empty parameters schema, and execute function that fetches and caches the complete llm.txt documentation from GitHub (~2000 lines).

  export const getLlmFullContextTool = {
    name: "get_llm_full_context",
    description: "Fetches the complete documentation bundle (~2000 lines, consumes considerable tokens but provides all documentation in one go). IMPORTANT: Ask the user for confirmation before executing this tool as it will add the entire Slice.js documentation to the context.",
    parameters: z.object({}),
    execute: async () => {
      const cached = getCached('llm.txt');
      if (cached) {
        console.error('[MCP] Returning cached llm.txt');
        return cached;
      }
  
      console.error('[MCP] Fetching llm.txt from GitHub');
      const url = `${BASE_URL}llm.txt`;
      try {
        const response = await fetch(url);
        if (!response.ok) throw new Error(`HTTP ${response.status}`);
        const content = await response.text();
        setCache('llm.txt', content);
        console.error('[MCP] Fetched and cached llm.txt, now populating individual doc cache');
  
        // Populate cache with individual docs from llm.txt
        const sections = content.split(/\n=== /).slice(1); // Skip first empty
        let populatedCount = 0;
        for (const section of sections) {
          const lines = section.split('\n');
          const filePath = lines[0].replace(' ===', ''); // e.g., 'markdown/getting-started.md'
          const docContent = lines.slice(1).join('\n').trim();
          if (filePath && docContent) {
            // Compute cache key as doc id: remove 'markdown/' prefix and '.md' suffix
            const docId = filePath.replace(/^markdown\//, '').replace(/\.md$/, '');
            setCache(docId, docContent);
            populatedCount++;
          }
        }
        console.error(`[MCP] Populated cache with ${populatedCount} individual docs`);
  
        // Update DOCS_STRUCTURE if already initialized
        if (isInitialized) {
          DOCS_STRUCTURE.length = 0; // Clear the array
          DOCS_STRUCTURE.push(...parseDocsFromLlmTxt(content));
          console.error(`[MCP] Updated DOCS_STRUCTURE to ${DOCS_STRUCTURE.length} documents from llm.txt`);
        }
  
        return content;
      } catch (error) {
        console.error(`[MCP] Error fetching llm.txt: ${error}`);
        return `Error fetching llm.txt: ${error}`;
      }
    },
  };

• src/tools/get-llm-full-context.ts:7-7 (schema)

  Schema definition using Zod with empty parameters object, indicating this tool takes no input parameters.

  parameters: z.object({}),

• src/index.ts:17-17 (registration)

  Tool registration where getLlmFullContextTool is added to the FastMCP server instance.

  server.addTool(getLlmFullContextTool);

• src/utils.ts:5-15 (helper)

  Cache retrieval helper function getCached used by the tool to check for cached llm.txt content.

  export function getCached(key: string): string | null {
    const cached = cache.get(key);
    if (!cached) return null;
  
    if (Date.now() - cached.timestamp > CACHE_TTL) {
      cache.delete(key);
      return null;
    }
  
    return cached.content;
  }

• src/utils.ts:17-19 (helper)

  Cache storage helper function setCache used by the tool to cache the fetched llm.txt content and individual documentation sections.

  export function setCache(key: string, content: string): void {
    cache.set(key, { content, timestamp: Date.now() });
  }