Merge pull request #8 from PatrickSys/feat/auto-heal-semantic-search

feat: Auto-heal for silent semantic search failure

Author: PatrickSys

CHANGELOG.md

@@ -1,6 +1,11 @@
 # Changelog
 
 
+## [1.3.1] - 2026-01-05
+
+### Fixed
+- **Auto-Heal Semantic Search**: Detects LanceDB schema corruption (missing `vector` column), triggers re-indexing, and retries search instead of silently falling back to keyword-only results.
+
 ## [1.3.0] - 2026-01-01
 
 ### Added

internal-docs

@@ -1,6 +1,6 @@
 {
   "name": "codebase-context",
-  "version": "1.3.0",
+  "version": "1.3.1",
   "description": "MCP server that helps AI agents understand your codebase - patterns, libraries, architecture, monorepo support",
   "type": "module",
   "main": "./dist/lib.js",
@@ -120,4 +120,4 @@
       "sharp"
     ]
   }
-}
+}

package.json

@@ -86,7 +86,7 @@ export class CodebaseIndexer {
       },
       embedding: {
         provider: 'transformers',
-        model: 'Xenova/bge-base-en-v1.5',
+        model: 'Xenova/bge-small-en-v1.5',
         batchSize: 100
       },
       skipEmbedding: false,

src/core/indexer.ts

@@ -9,6 +9,7 @@ import { CodeChunk, SearchResult, SearchFilters } from '../types/index.js';
 import { EmbeddingProvider, getEmbeddingProvider } from '../embeddings/index.js';
 import { VectorStorageProvider, getStorageProvider } from '../storage/index.js';
 import { analyzerRegistry } from './analyzer-registry.js';
+import { IndexCorruptedError } from '../errors/index.js';
 
 export interface SearchOptions {
   useSemanticSearch?: boolean;
@@ -62,6 +63,9 @@ export class CodebaseSearcher {
 
       this.initialized = true;
     } catch (error) {
+      if (error instanceof IndexCorruptedError) {
+        throw error; // Propagate to handler for auto-heal
+      }
       console.warn('Partial initialization (keyword search only):', error);
       this.initialized = true;
     }
@@ -217,6 +221,9 @@ export class CodebaseSearcher {
           }
         });
       } catch (error) {
+        if (error instanceof IndexCorruptedError) {
+          throw error; // Propagate to handler for auto-heal
+        }
         console.warn('Semantic search failed:', error);
       }
     }

src/core/search.ts

@@ -0,0 +1,10 @@
+/**
+ * Thrown when the LanceDB index is corrupted or has a schema mismatch.
+ * This error signals that re-indexing is required for semantic search to work.
+ */
+export class IndexCorruptedError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = 'IndexCorruptedError';
+  }
+}

src/errors/index.ts

@@ -20,12 +20,13 @@ import {
   Resource
 } from '@modelcontextprotocol/sdk/types.js';
 import { CodebaseIndexer } from './core/indexer.js';
-import { IndexingStats } from './types/index.js';
+import { IndexingStats, SearchResult } from './types/index.js';
 import { CodebaseSearcher } from './core/search.js';
 import { analyzerRegistry } from './core/analyzer-registry.js';
 import { AngularAnalyzer } from './analyzers/angular/index.js';
 import { GenericAnalyzer } from './analyzers/generic/index.js';
 import { InternalFileGraph } from './utils/usage-tracker.js';
+import { IndexCorruptedError } from './errors/index.js';
 
 analyzerRegistry.register(new AngularAnalyzer());
 analyzerRegistry.register(new GenericAnalyzer());
@@ -62,11 +63,10 @@ const indexState: IndexState = {
   status: 'idle'
 };
 
-
 const server: Server = new Server(
   {
     name: 'codebase-context',
-    version: '1.3.0'
+    version: '1.3.1'
   },
   {
     capabilities: {
@@ -492,7 +492,62 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
         }
 
         const searcher = new CodebaseSearcher(ROOT_PATH);
-        const results = await searcher.search(query, limit || 5, filters);
+        let results: SearchResult[];
+
+        try {
+          results = await searcher.search(query, limit || 5, filters);
+        } catch (error) {
+          if (error instanceof IndexCorruptedError) {
+            console.error('[Auto-Heal] Index corrupted. Triggering full re-index...');
+
+            await performIndexing();
+
+            if (indexState.status === 'ready') {
+              console.error('[Auto-Heal] Success. Retrying search...');
+              const freshSearcher = new CodebaseSearcher(ROOT_PATH);
+              try {
+                results = await freshSearcher.search(query, limit || 5, filters);
+              } catch (retryError) {
+                return {
+                  content: [
+                    {
+                      type: 'text',
+                      text: JSON.stringify(
+                        {
+                          status: 'error',
+                          message: `Auto-heal retry failed: ${
+                            retryError instanceof Error ? retryError.message : String(retryError)
+                          }`
+                        },
+                        null,
+                        2
+                      )
+                    }
+                  ]
+                };
+              }
+            } else {
+              return {
+                content: [
+                  {
+                    type: 'text',
+                    text: JSON.stringify(
+                      {
+                        status: 'error',
+                        message: `Auto-heal failed: Indexing ended with status '${indexState.status}'`,
+                        error: indexState.error
+                      },
+                      null,
+                      2
+                    )
+                  }
+                ]
+              };
+            }
+          } else {
+            throw error; // Propagate unexpected errors
+          }
+        }
 
         return {
           content: [
@@ -538,19 +593,19 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
                   lastIndexed: indexState.lastIndexed?.toISOString(),
                   stats: indexState.stats
                     ? {
-                      totalFiles: indexState.stats.totalFiles,
-                      indexedFiles: indexState.stats.indexedFiles,
-                      totalChunks: indexState.stats.totalChunks,
-                      duration: `${(indexState.stats.duration / 1000).toFixed(2)}s`
-                    }
+                        totalFiles: indexState.stats.totalFiles,
+                        indexedFiles: indexState.stats.indexedFiles,
+                        totalChunks: indexState.stats.totalChunks,
+                        duration: `${(indexState.stats.duration / 1000).toFixed(2)}s`
+                      }
                     : undefined,
                   progress: progress
                     ? {
-                      phase: progress.phase,
-                      percentage: progress.percentage,
-                      filesProcessed: progress.filesProcessed,
-                      totalFiles: progress.totalFiles
-                    }
+                        phase: progress.phase,
+                        percentage: progress.percentage,
+                        filesProcessed: progress.filesProcessed,
+                        totalFiles: progress.totalFiles
+                      }
                     : undefined,
                   error: indexState.error,
                   hint: 'Use refresh_index to manually trigger re-indexing when needed.'

src/index.ts

@@ -6,6 +6,7 @@
 import { promises as fs } from 'fs';
 import { VectorStorageProvider, CodeChunkWithEmbedding, VectorSearchResult } from './types.js';
 import { CodeChunk, SearchFilters } from '../types/index.js';
+import { IndexCorruptedError } from '../errors/index.js';
 
 export class LanceDBStorageProvider implements VectorStorageProvider {
   readonly name = 'lancedb';
@@ -20,44 +21,33 @@ export class LanceDBStorageProvider implements VectorStorageProvider {
 
     try {
       this.storagePath = storagePath;
-
-      // Ensure directory exists
       await fs.mkdir(storagePath, { recursive: true });
 
-      // Dynamic import to avoid issues at require time
       const lancedb = await import('@lancedb/lancedb');
-
-      // Connect to database
       this.db = await lancedb.connect(storagePath);
 
-      // Check if table exists and has valid schema
+      // Check if table exists and validate schema
       const tableNames = await this.db.tableNames();
       if (tableNames.includes('code_chunks')) {
         this.table = await this.db.openTable('code_chunks');
 
-        // Validate schema has vector column (required for semantic search)
-        try {
-          const schema = await this.table.schema();
-          const hasVectorColumn = schema.fields.some((f: any) => f.name === 'vector');
-
-          if (!hasVectorColumn) {
-            console.error('Stale index detected (missing vector column). Rebuilding...');
-            await this.db.dropTable('code_chunks');
-            this.table = null;
-          } else {
-            console.error('Opened existing LanceDB table');
-          }
-        } catch (_schemaError) {
-          // If schema check fails, table is likely corrupted - drop and rebuild
-          console.error('Failed to validate table schema, rebuilding index...');
-          await this.db.dropTable('code_chunks');
-          this.table = null;
+        const schema = await this.table.schema();
+        const hasVectorColumn = schema.fields.some((f: any) => f.name === 'vector');
+
+        if (!hasVectorColumn) {
+          throw new IndexCorruptedError('LanceDB index corrupted: missing vector column');
         }
+        console.error('Opened existing LanceDB table');
+      } else {
+        this.table = null;
       }
 
       this.initialized = true;
       console.error(`LanceDB initialized at: ${storagePath}`);
     } catch (error) {
+      if (error instanceof IndexCorruptedError) {
+        throw error;
+      }
       console.error('Failed to initialize LanceDB:', error);
       throw error;
     }
@@ -115,7 +105,7 @@ export class LanceDBStorageProvider implements VectorStorageProvider {
     filters?: SearchFilters
   ): Promise<VectorSearchResult[]> {
     if (!this.initialized || !this.table) {
-      return [];
+      throw new IndexCorruptedError('LanceDB index corrupted: no table available for search');
     }
 
     try {
@@ -170,7 +160,13 @@ export class LanceDBStorageProvider implements VectorStorageProvider {
         distance: result._distance || 0
       }));
     } catch (error) {
-      console.error('Failed to search:', error);
+      // Only trigger auto-heal for verified corruption patterns
+      if (error instanceof Error && error.message.toLowerCase().includes('no vector column')) {
+        throw new IndexCorruptedError(`LanceDB index corrupted: ${error.message}`);
+      }
+
+      // Transient errors - log and gracefully degrade
+      console.error('[LanceDB] Search error:', error instanceof Error ? error.message : error);
       return [];
     }
   }
@@ -199,8 +195,7 @@ export class LanceDBStorageProvider implements VectorStorageProvider {
     }
 
     try {
-      const result = await this.table.countRows();
-      return result;
+      return await this.table.countRows();
     } catch (error) {
       console.error('Failed to count rows:', error);
       return 0;