Obsidian Diary MCP Server

AI-powered journaling with local processing, automatic backlinks, and smart prompts.

Features

• AI-generated reflection prompts based on past 3 calendar days

• Day citations with automatic [[YYYY-MM-DD]] backlinks

• Brain dump prioritization (analyzes your writing, not prompts)

• Smart #tag extraction using theme similarity

• Todo extraction to organized checklists

• Memory trace analysis with theme evolution

• Sunday synthesis (weekly reflection prompts)

Requirements

• uv (Python package manager)

• Ollama (llama3.1 or compatible model)

• MCP client (e.g., GitHub Copilot CLI)

• Obsidian vault (for markdown files)

Setup

1. Clone and install:

git clone https://github.com/madebygps/obsidian-diary-mcp.git

cd obsidian-diary-mcp

uv sync

chmod +x start-server.sh

2. Configure:

cp .env.example .env

Edit .env: set DIARY_PATH and PLANNER_PATH (required)

3. Add to MCP client config (e.g., GitHub Copilot CLI):

• Name: diary

• Command: /full/path/to/obsidian-diary-mcp/start-server.sh

Configuration (.env):

Required: DIARY_PATH, PLANNER_PATH

Optional: OLLAMA_MODEL (default: llama3.1:latest), OLLAMA_TIMEOUT (60s), OLLAMA_TEMPERATURE (0.7), OLLAMA_NUM_PREDICT (1000 tokens)

Usage

1. Create: "create a memory log for today" → AI prompts based on past 3 days

2. Write: Open in Obsidian, write freely in Brain Dump section

3. Extract: "extract todos from today's entry" → Action items to planner

4. Link: "link today's memory log" → Auto-generates [[YYYY-MM-DD]] & #tags

5. Explore: Use Obsidian's backlinks panel and graph view

More Commands: "show themes from last week", "create memory trace for 30 days", "refresh memory links for 30 days"

Debugging

Logs in logs/ directory: server-YYYY-MM-DD.log (protocol), debug-YYYY-MM-DD.log (operations)

tail -f logs/debug-$(date +%Y-%m-%d).log  # Watch in real-time
grep ERROR logs/debug-*.log                # Find errors
grep "similarity" logs/debug-*.log         # Debug backlinks

Troubleshooting

Server issues: Check .env exists with DIARY_PATH and PLANNER_PATH set. Run ./start-server.sh directly to test.

Ollama issues: Verify running with curl http://localhost:11434/api/tags. Pull model: ollama pull llama3.1:latest

No backlinks: Need 2+ entries with similar themes (>8% overlap). Ensure Brain Dump section has substantial content (>50 chars). Check: grep "Brain Dump" logs/debug-*.log

Timeouts: Increase OLLAMA_TIMEOUT (90+) and OLLAMA_NUM_PREDICT (2000+) for reasoning models.

How It Works

• Local AI: Ollama processes entries locally—content never leaves your machine

• Calendar-Based: Analyzes past 3 calendar days (not just last 3 entries)

• Brain Dump Focus: Prioritizes your writing over answered prompts for themes

• Day Citations: AI cites [Day 1]/[Day 2] → converts to [[2025-10-07]] backlinks

• Smart Linking: Jaccard similarity connects entries with >8% theme overlap

• Sundays: 5 weekly synthesis prompts (vs 3 daily)

• Todo Extraction: AI identifies action items from brain dumps

Entry Format

Each entry (YYYY-MM-DD.md) has plain text headers:

## Reflection Prompts
**1. Question with [[2025-10-06]] backlink (reason)...**

---

## Brain Dump
Your thoughts, experiences, observations...

---

## Memory Links
**Temporal connections:** [[2025-10-05]] • [[2025-10-04]]
**Topic tags:** #career-growth #self-reflection

License

MIT • Python 3.13+ • FastMCP 2.12.4+ • Ollama