# knowledge-base MCP server

> Provides AI agents with persistent memory by capturing, classifying, and retrieving knowledge from Obsidian, code, and web sources.

- Repository: https://github.com/willynikes2/knowledge-base-server
- Page: https://hostedmcp.org/servers/willynikes2/knowledge-base-server
- Transport: —
- Stars: 190
- License: MIT

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `KB_API_KEY_CLAUDE` | no | no |  |
| `KB_API_KEY_GEMINI` | no | no |  |
| `KB_API_KEY_OPENAI` | no | no |  |
| `KB_PASSWORD` | no | no |  |
| `OBSIDIAN_VAULT_PATH` | no | no |  |

## Tools (21)

- `kb_search`
- `kb_search_smart`
- `kb_context`
- `kb_read`
- `kb_list`
- `kb_write`
- `kb_ingest`
- `kb_classify`
- `kb_promote`
- `kb_synthesize`
- `kb_capture_session`
- `kb_capture_fix`
- `kb_capture_web`
- `kb_capture_youtube`
- `kb_vault_status`
- `kb_safety_check`
- `bus_send`
- `bus_inbox`
- `bus_wait`
- `kb_mistake_record`
- `kb_patterns_for`

## FAQ

### How does the server save on token usage?

The kb_context tool returns summaries and metadata instead of full content, resulting in over 90% token savings.

### Does this require an external database or cloud dependency?

No, it uses SQLite FTS5 and requires no external databases, Docker, or cloud dependencies.

### Can I use semantic search without an API key?

Yes, semantic and vector search is performed via local HuggingFace embeddings, so no API keys are needed for that feature.

### What kind of files can be ingested?

The system supports over 20 file types, including Markdown, text, code, PDFs, JSON, YAML, XML, and CSV.