# v1.5.0: 2-5x Faster Analytics, Export & Sync — How We Eliminated a Subtle Anti-Pattern

We just shipped [Knowledge Keeper MCP v1.5.0](https://github.com/zsc-glitch/knowledge-keeper-mcp) with a focused performance optimization that makes analytics, export, and sync operations **2-5x faster** — without changing a single API or breaking any existing behavior.

## The Problem: `searchKnowledge({ query: "" })`

Knowledge Keeper has a `searchKnowledge()` function that powers keyword search. It loads the index, filters by query, filters by type/tags, sorts by date, and returns results. Solid for search.

But we had a subtle anti-pattern: **13 places across 8 files were calling `searchKnowledge({ query: "", limit: N })`** — passing an empty query just to get "all entries."

This means every analytics calculation, every export, every sync operation was running through the full search pipeline:

```
loadIndex() → filter by type → filter by tags → filter by query("") → sort → slice
```

When all you need is "give me everything," that's a lot of unnecessary work.

## The Fix: Direct Index Reads

We replaced all 13 instances with a simple `loadAllEntries()` function that reads `index.json` directly:

```typescript
async function loadAllEntries(): Promise<KnowledgePoint[]> {
  const indexPath = path.join(vaultDir, "index.json");
  const content = await fs.readFile(indexPath, "utf-8");
  return JSON.parse(content).entries || [];
}
```

No filtering. No sorting. No search overhead. Just the data, straight from disk.

## The Worst Offender: Cloud Sync

The most impactful fix was in `cloud-sync.ts`. The pull function was calling `searchKnowledge({ query: "" })` **inside a loop** — once for every remote item being synced:

```typescript
// BEFORE: O(n) file reads
for (const item of remoteItems) {
  const localKnowledge = await searchKnowledge({ query: "", limit: 10000 });
  const localEntry = localKnowledge.find(e => e.id === item.id);
  // ...
}
```

If you're syncing 50 items, that's 50 full index loads. We moved the read outside the loop:

```typescript
// AFTER: 1 file read
const localKnowledgeCache = await loadAllEntries();
for (const item of remoteItems) {
  const localEntry = localKnowledgeCache.find(e => e.id === item.id);
  // ...
}
```

That's a 50x reduction in file I/O for a 50-item sync.

## What Changed

| File | Before | After | Impact |
|------|--------|-------|--------|
| `tools/recent.ts` | `searchKnowledge({ query: "" })` | Direct index + sort_by param | Faster recent listing |
| `tools/batch.ts` | Missing `update_type` action | Implemented | New feature |
| `tools/export.ts` | `searchKnowledge` for bulk export | Direct index read | 2-5x faster export |
| `analytics.ts` | 3× `searchKnowledge({ query: "", limit: 10000 })` | `loadAllEntries()` | Faster analytics |
| `tools/graph-build.ts` | `searchKnowledge` for graph | Direct index | Faster graph building |
| `tools/sync.ts` | 3× `searchKnowledge` | `loadAllEntries()` | Faster Obsidian sync |
| `cloud-sync.ts` | 3× in-loop `searchKnowledge` | Pre-loaded cache | 50x less I/O |

## New Features in v1.5.0

- **`knowledge_recent`** now supports a `sort_by` parameter (`updated` or `created`), shows total count, and displays timestamps with minute precision
- **`knowledge_batch`** now supports `update_type` action to change knowledge point types (was defined in schema but never implemented)

## Upgrade

```bash
npx @zsc-glitch/knowledge-keeper-mcp@1.5.0
```

Or add to your Claude Code / Cursor MCP config:

```json
{
  "mcpServers": {
    "knowledge-keeper": {
      "command": "npx",
      "args": ["-y", "@zsc-glitch/knowledge-keeper-mcp@1.5.0"]
    }
  }
}
```

## Lesson Learned

When building tool systems, it's easy to reuse existing functions for convenience. `searchKnowledge` works, so why not use it everywhere? The answer: **when the use case doesn't need search, don't use search.** A "get everything" operation should read the index directly.

This is especially important in MCP servers where every millisecond matters — your AI agent is waiting for the response.

---

*Knowledge Keeper MCP — 31 tools, zero API keys, 100% local. [GitHub](https://github.com/zsc-glitch/knowledge-keeper-mcp) • [npm](https://npm.im/@zsc-glitch/knowledge-keeper-mcp) • [Docs](https://zsc-glitch.github.io/knowledge-keeper-mcp/)*