@coderule/mcp 1.6.3 → 1.7.0

package/dist/mcp-cli.js

@@ -1046,6 +1046,11 @@ async function walkDirectory(current, opts, stats) {
         stats.skipped += 1;
         continue;
       }
+      if (stat.size === 0) {
+        stats.skipped += 1;
+        dirLogger.debug({ relPath }, "Skipping zero-length file");
+        continue;
+      }
       const target = dirent.isSymbolicLink() ? await readSymlinkTarget(absPath, dirLogger) : null;
       const state = opts.filesRepo.upsertFromStat({
         relPath,
@@ -1564,6 +1569,13 @@ var ServiceRunner = class {
       this.runtime.logger.debug({ relPath }, "Watcher event ignored by rules");
       return;
     }
+    if (fileStats.size === 0) {
+      this.runtime.logger.debug(
+        { relPath },
+        "Watcher skipping zero-length file"
+      );
+      return;
+    }
     const isSymlink = fileStats.isSymbolicLink();
     const target = isSymlink ? await readSymlinkTarget2(absPath) : null;
     const state = this.runtime.filesRepo.upsertFromStat({
@@ -1916,9 +1928,21 @@ function createMcpServer({
   server.registerTool(
     "check",
     {
-      title: "Indexer status",
-      description: "Inspect the current indexing state, snapshot, and queue metrics",
-      inputSchema: {}
+      title: "Coderule Indexing Status",
+      description: `Check the current state of the code indexer. Returns repository info, file states (clean/dirty/hashing/missing), queue metrics, latest snapshot details, and service health.
+
+Use this tool to:
+- Debug why queries are failing or returning no results
+- Verify the indexer is ready before querying
+- Check which repository is currently indexed
+- Monitor indexing progress`,
+      inputSchema: {},
+      annotations: {
+        readOnlyHint: true,
+        destructiveHint: false,
+        idempotentHint: true,
+        openWorldHint: false
+      }
     },
     async () => {
       const status = await collectIndexingStatus(runtime, runner);
@@ -1928,16 +1952,51 @@ function createMcpServer({
       };
     }
   );
-  const queryInputSchema = {
-    query: z.string().min(1, "Query text is required"),
-    budgetTokens: z.number().int().positive().optional().describe("Token budget for retrieval (default 3000)")
-  };
   server.registerTool(
     "query",
     {
-      title: "Snapshot retrieval query",
-      description: "Execute a retrieval query against the most recent indexed snapshot",
-      inputSchema: queryInputSchema
+      title: "Semantic Code Retrieval",
+      description: `Graph-based RAG for semantic code search across the indexed codebase.
+
+\u{1F6A8} CRITICAL: Use this tool FIRST before any code modification, bug fix, feature addition, or architecture exploration.
+
+QUERY STRATEGY:
+Use keyword-rich technical terms, NOT natural language questions.
+
+\u2705 GOOD: "JWT authentication middleware token validation security handler FastAPI"
+\u2705 GOOD: "file indexing worker celery queue job AST parser embedding chunker"
+\u2705 GOOD: "database schema SQLAlchemy model migration table column relationship"
+
+\u274C BAD: "How does authentication work?"
+\u274C BAD: "Show me the login code"
+
+RETRIEVAL MECHANISM:
+- Seeds: Semantic similarity via embeddings (~6 initial matches)
+- Flood: Graph propagation through AST relationships (imports, calls, inheritance)
+- Neighbors: Adjacent code chunks for context
+- Calls: Function call relationships
+
+Returns 30-50 code segments with file paths, line numbers, and relevance scores.
+
+DEFAULT BUDGET: 10,000 tokens (sufficient for most tasks)
+
+WHEN TO USE:
+- Before modifying ANY existing code
+- Understanding system architecture
+- Finding related implementations
+- Locating tests or examples
+- Bug investigation
+- Feature implementation planning`,
+      inputSchema: {
+        query: z.string().min(1, "Query text is required"),
+        budgetTokens: z.number().int().positive().optional().describe("Token budget for retrieval (default 10000)")
+      },
+      annotations: {
+        readOnlyHint: true,
+        destructiveHint: false,
+        idempotentHint: true,
+        openWorldHint: false
+      }
     },
     async ({
       query,
@@ -1949,7 +2008,8 @@ function createMcpServer({
         const statusText = formatStatus(
           await collectIndexingStatus(runtime, runner)
         );
-        const text = `We are not ready....
+        const text = `Indexer not ready. Current status:
+
 ${statusText}`;
         return { content: [{ type: "text", text }] };
       }
@@ -1961,11 +2021,12 @@ ${statusText}`;
         const statusText = formatStatus(
           await collectIndexingStatus(runtime, runner)
         );
-        const text = `We are not ready....
+        const text = `Snapshot not ready on server. Current status:
+
 ${statusText}`;
         return { content: [{ type: "text", text }] };
       }
-      const effectiveBudget = Math.max(100, budgetTokens ?? 3e3);
+      const effectiveBudget = Math.max(100, budgetTokens ?? 1e4);
       try {
         const result = await runtime.clients.retrieval.query(
           readyHash,