promptgraph-mcp 2.9.12 → 2.9.14

package/index.js

@@ -106,7 +106,7 @@ if (COMMAND_MAP[args[0]]) {
 // ── MCP server mode (no CLI command) ──
 const { Server } = await import('@modelcontextprotocol/sdk/server/index.js');
 const { StdioServerTransport } = await import('@modelcontextprotocol/sdk/server/stdio.js');
-const { CallToolRequestSchema, ListToolsRequestSchema } = await import('@modelcontextprotocol/sdk/types.js');
+const { CallToolRequestSchema, ListToolsRequestSchema, ListPromptsRequestSchema, GetPromptRequestSchema } = await import('@modelcontextprotocol/sdk/types.js');
 const { search, getContext, getCallers, getCallees, getImpact, listAll } = await import('./search.js');
 const { loadConfig: _loadConfig, saveConfig: _saveConfig } = await import('./config.js');
 const { startWatcher } = await import('./watcher.js');
@@ -116,7 +116,7 @@ const { createRequire } = await import('module');
 const pkg = createRequire(import.meta.url)('./package.json');
 const server = new Server(
   { name: 'promptgraph', version: pkg.version },
-  { capabilities: { tools: {} } }
+  { capabilities: { tools: {}, prompts: {} } }
 );
 
 server.setRequestHandler(ListToolsRequestSchema, async () => ({
@@ -332,6 +332,46 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
   }
 });
 
+server.setRequestHandler(ListPromptsRequestSchema, async () => ({
+  prompts: [
+    {
+      name: 'pg',
+      description: 'Search skills and load the best match for your task',
+      arguments: [{ name: 'query', description: 'What you want to do', required: true }],
+    },
+    {
+      name: 'pg-list',
+      description: 'List all indexed skills',
+      arguments: [],
+    },
+  ],
+}));
+
+server.setRequestHandler(GetPromptRequestSchema, async (request) => {
+  const { name, arguments: args } = request.params;
+  if (name === 'pg') {
+    const query = args?.query || '';
+    const results = await search(query, 3);
+    if (!results.length) return { messages: [{ role: 'user', content: { type: 'text', text: `No skills found for: ${query}` } }] };
+    const top = results[0];
+    const fs = await import('fs');
+    let content = '';
+    try { content = fs.readFileSync(top.path, 'utf8'); } catch {}
+    return {
+      messages: [{
+        role: 'user',
+        content: { type: 'text', text: `# Skill: ${top.name} (score: ${top.score?.toFixed(2)})\n\n${content}` },
+      }],
+    };
+  }
+  if (name === 'pg-list') {
+    const skills = await listAll();
+    const text = skills.map(s => `- **${s.name}** — ${s.description || ''}`).join('\n');
+    return { messages: [{ role: 'user', content: { type: 'text', text: text || 'No skills indexed.' } }] };
+  }
+  throw new Error(`Unknown prompt: ${name}`);
+});
+
 startWatcher();
 
 const transport = new StdioServerTransport();

package/indexer.js

@@ -53,15 +53,7 @@ export async function indexBatch(db, skills, { fast = false } = {}) {
     }
   }
 
-  let embeddings = [];
-  if (!fast && allChunks.length) {
-    const texts = allChunks.map(c => c.text);
-    process.stdout.write(`  Embedding ${texts.length} chunks...`);
-    embeddings = await embedBatch(texts);
-    process.stdout.write('\r' + ' '.repeat(40) + '\r');
-  }
-
-  // pass 1: upsert all skills + chunks (no edges yet)
+  // pass 1: upsert all skills metadata first (so chunks can reference them)
   db.transaction(() => {
     for (const skill of skills) {
       const id = skillId(skill.source, skill.name);
@@ -72,14 +64,29 @@ export async function indexBatch(db, skills, { fast = false } = {}) {
         deleteEdges.run(id);
       }
     }
-    if (!fast) {
-      for (let i = 0; i < allChunks.length; i++) {
-        const { id, chunkIndex, text } = allChunks[i];
-        upsertChunk.run(id, chunkIndex, text, vecToBlob(embeddings[i]));
-      }
-    }
   })();
 
+  // pass 1b: embed in batches of 50 with progress bar, save each batch immediately
+  if (!fast && allChunks.length) {
+    const EMBED_BATCH = 50;
+    const total = allChunks.length;
+    for (let i = 0; i < total; i += EMBED_BATCH) {
+      const batch = allChunks.slice(i, i + EMBED_BATCH);
+      const texts = batch.map(c => c.text);
+      const vecs = await embedBatch(texts);
+      db.transaction(() => {
+        for (let j = 0; j < batch.length; j++) {
+          const { id, chunkIndex, text } = batch[j];
+          upsertChunk.run(id, chunkIndex, text, vecToBlob(vecs[j]));
+        }
+      })();
+      const done = Math.min(i + EMBED_BATCH, total);
+      const pct = Math.round(done / total * 100);
+      process.stdout.write(`\r  Embedding chunks... ${done}/${total} (${pct}%)   `);
+    }
+    process.stdout.write('\r' + ' '.repeat(50) + '\r');
+  }
+
   // pass 2: resolve edges after all skills in batch are committed
   const resolveSameSource = db.prepare("SELECT id FROM skills WHERE name = ? AND source = ? LIMIT 1");
   const resolveAny = db.prepare("SELECT id FROM skills WHERE name = ? ORDER BY id LIMIT 1");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "promptgraph-mcp",
-  "version": "2.9.12",
+  "version": "2.9.14",
   "files": [
     "*.js",
     "commands/",