second-brain MCP Server

second-brain MCP Server

A self-maintaining personal knowledge database — powered by MCP, DuckDB, and biological memory models.

For anyone who saves more papers, notes, and figures than they could ever re-read.second-brain turns everything you capture into a database that maintains itself — auto-linking related notes, compressing what you stop reading, and keeping every figure searchable by its content. What you saved a year ago is still one query away, at a fraction of the token cost.

Why Does This Exist?

The One-Command Demo

save_article("https://arxiv.org/abs/2405.01234")
  ↓
• /abs/ auto-upgraded to /html/ — full paper, not just abstract
• Full text converted to Markdown
• All figures downloaded + OCR'd by Claude Vision
• Semantic embeddings computed
• Auto-linked to related notes already in your vault   ← auto-wikilinks
• Stored in 30-resources/ — queryable immediately

search_figures("UMAP cluster batch correction")
  ↓
• Returns the exact figure from the exact paper
• Works across your entire saved literature library

What Makes It Different

flowchart LR
    subgraph input["📥 Any Content Source"]
        A1["arXiv / PubMed paper"]
        A2["Web article / blog"]
        A3["Local PDF / DOCX"]
        A4["Personal note"]
    end

    subgraph core["⚙️ second-brain-mcp"]
        B1["Markdown note<br/>30-resources/"]
        B2["Figure OCR<br/>+ VLM description"]
        B3["Semantic embedding<br/>+ auto-wikilinks"]
        B4["Ebbinghaus score<br/>ranking"]
        B5["PNG snapshots<br/>60–90% token reduction"]
    end

    subgraph query["🔍 Queryable Knowledge"]
        C1["search_figures<br/>'UMAP melanocyte'"]
        C2["search_notes<br/>'batch correction scRNA'"]
        C3["get_context<br/>top-20 relevant notes"]
    end

    input --> core
    B1 --> B2
    B1 --> B3
    B3 --> B4
    B4 --> B5
    B2 --> C1
    B3 --> C2
    B4 --> C3

Eight things most self-hosted memory tools can't do — combined in one:

Cross-Session Continuity — Pick Up Where You Left Off

Every project you work on can be resumed in a new session with full context — no re-explaining, no lost progress.

flowchart LR
    A["🟢 Session Start<br/>get_context()"] --> B["AI receives:<br/>• goals.md — current priorities<br/>• Top-20 recent notes<br/>• Extracted rules"]
    B --> C["Work on project<br/>new_note / search / read"]
    C --> D["🔴 Before ending session<br/>update_goals(...)"]
    D --> E["New session<br/>get_context() again"]
    E --> B

How It Works in Practice

End of session — tell the agent to save state:

Update goals: currently working on the scRNA batch correction pipeline.
Completed: harmony integration. Blocked on: choosing n_components for PCA.
Next session: start from the PCA parameter sweep in 20-areas/research/harmony-notes.md

The agent calls update_goals() and optionally new_note("project", ...) for detailed progress.

Start of next session — just say:

Get context and continue where we left off.

The agent calls get_context() and immediately sees:

• goals.md with the state you saved
• The harmony-notes.md surfaced at the top (recently accessed, high Ebbinghaus score)
• Rules auto-extracted from that note, e.g.:

RULE: use n_components=30 for this dataset — tested 20/30/50, 30 minimises batch effect without losing resolution
RULE: exclude sample CRC_04 — library size outlier confirmed by QC

These rules live in memory/rules.md and are injected at every get_context() call — the AI carries your hard-won decisions forward automatically, without you having to repeat them.

What Gets Persisted

This works across any project — bioinformatics analysis, coding, writing, research. Save state with one sentence at the end of a session; resume instantly at the start of the next.

Example Queries

# Resume a project from last session
get_context()  # → goals + recent notes + rules loaded automatically

# Find a specific figure panel across all saved papers
search_figures("p < 0.001 UMAP cluster")

# Semantic search across all notes
search_notes("single cell integration batch correction")

# Decision records for a specific project
get_decisions("MyProject")

Memory Architecture — Biological Analogy

Token Efficiency

Memory that gets cheaper over time — unlike flat-file systems where old notes cost the same forever.

Note age →   fresh (0–3 mo)   3–6 months     6–12 months    1 year+
             ──────────────   ──────────     ───────────    ───────
token cost:  ██████████████   ██████         ████           ██
             ~1,000 tokens    ~400 tokens    ~256 tokens    ~100 tokens
                              ▼ 60%          ▼ 74%          ▼ 90%

Tier assigned by score × age (adaptive). Frequently-accessed notes stay full-text regardless of age.

Search Performance

Measured on Apple Silicon MacBook (20-rep average, BM25-only mode).

Vault    BM25-only p50          Hybrid BM25+semantic p50
──────   ─────────────────      ────────────────────────
10 n     ████░░░░░   21 ms      ████████████   37 ms
50 n     ██████░░░   25 ms      █████████████  39 ms
100 n    ███████░░   27 ms      ██████████████ 45 ms

Hybrid mode adds ~18 ms for embedding lookup. Both modes scale sub-linearly with vault size.

Recall figures at this scale (10–100 notes) carry high sample variance — a single ambiguous query shifts Recall@1 by 10%. Treat them as directional, not as benchmarks against large corpora; the takeaway is that hybrid consistently beats BM25-only on relevance for a fixed query set.

System Architecture

┌─────────────────────────────────────────────────────┐
│                    AI Agent Layer                    │
│         Claude Code · Gemini CLI · Any MCP           │
└──────────────────────┬──────────────────────────────┘
                       │ MCP Protocol (27 tools)
┌──────────────────────▼──────────────────────────────┐
│               Layer 2 — MCP Server                   │
│                    server.py                         │
│  get_context · search_notes · save_article · … (27)  │
└──────┬───────────────┬────────────────┬─────────────┘
       │               │                │
┌──────▼──────┐ ┌──────▼──────┐ ┌──────▼──────┐
│  vault_sleep│ │  vault_db   │ │  figures    │
│  compress   │ │  DuckDB FTS │ │  PNG snap   │
│  Phase 3–9  │ │  + semantic │ │  OCR · VLM  │
└──────┬──────┘ └──────┬──────┘ └─────────────┘
       │               │
┌──────▼───────────────▼──────────────────────────────┐
│               Layer 0 — Markdown Vault               │
│   00-inbox · 10-projects · 20-areas · 30-resources   │
│   40-archive · decisions · memory · templates        │
│         (syncs via Google Drive / iCloud / git)      │
└─────────────────────────────────────────────────────┘

Vault Sleep — Auto-compression Flow

Every Sunday 02:00 (launchd, no interaction needed)
        │
        ▼
 sync_index + embeddings
        │
        ▼  age > 90d AND Ebbinghaus score ≤ 0.5
 ┌──────────────────────────────────────┐
 │         Adaptive Tier Selection      │
 │  score > 1.5  →  text  (keep full)  │  ← frequently-read: never compressed
 │  score > 0.8  →  large  ~400 tokens │
 │  score > 0.3  →  base   ~256 tokens │
 │  otherwise    →  small  ~100 tokens │
 └────────────────┬─────────────────────┘
                  │
  Gemini CLI → Claude CLI → naive   (auto-fallback, no LLM required)
                  │
    compressed → vault  /  original → 40-archive/  /  snapshot → .png

MCP Tools (27 total)

Test Results

tests/test_figures.py      19 passed   (OCR, snapshots, VLM)
tests/test_server.py       13 passed   (MCP tools, path safety)
tests/test_vault_db.py     39 passed   (FTS, semantic search, embeddings)
tests/test_vault_sleep.py  44 passed   (compression, consolidation, rules, prune)
────────────────────────────────────────
115 passed in 3.37s

Installation

Prerequisites

Vault structure is auto-created on first server start — no manual mkdir needed.

macOS / Linux — Quick Start

Step 1 — Install

pip install mcp-second-brain
playwright install chromium

Step 2 — Register with your AI agent

Claude Code (CLI) — global, works in any project:

claude mcp add --scope user second-brain \
  --env SECOND_BRAIN_PATH=~/second-brain \
  -- python -m mcp_second_brain

Claude Desktop — add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "second-brain": {
      "command": "python",
      "args": ["-m", "mcp_second_brain"],
      "env": { "SECOND_BRAIN_PATH": "/Users/yourname/second-brain" }
    }
  }
}

Step 3 — First run

Start your agent and say:

init_vault

The server auto-creates all directories and templates on startup. Call init_vault explicitly to verify or repair the structure.

Windows — Quick Start

Step 1 — Install Python and the package

# Install Python 3.11+ from python.org, or via winget:
winget install Python.Python.3.11

pip install mcp-second-brain
playwright install chromium

Step 2 — Choose a vault location

# Local folder:
$vault = "C:\Users\$env:USERNAME\second-brain"

# Or a cloud-synced folder (Google Drive, OneDrive, etc.):
$vault = "G:\My Drive\second-brain"

The vault directories and templates are created automatically when the server first starts. No manual setup needed.

Step 3 — Register with your AI agent

Claude Code (VSCode extension) — create .mcp.json in your vault folder:

{
  "mcpServers": {
    "second-brain": {
      "command": "python",
      "args": ["-m", "mcp_second_brain"],
      "env": { "SECOND_BRAIN_PATH": "C:\\Users\\YourName\\second-brain" }
    }
  }
}

Claude Desktop — add to %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "second-brain": {
      "command": "python",
      "args": ["-m", "mcp_second_brain"],
      "env": { "SECOND_BRAIN_PATH": "C:\\Users\\YourName\\second-brain" }
    }
  }
}

Gemini CLI / other IDEs — edit %USERPROFILE%\.gemini\mcp_config.json (or IDE-specific config):

{
  "mcpServers": {
    "second-brain": {
      "command": "C:\\Users\\YourName\\.venvs\\mcp-second-brain\\Scripts\\python.exe",
      "args": ["-m", "mcp_second_brain"],
      "env": { "SECOND_BRAIN_PATH": "C:\\Users\\YourName\\second-brain" }
    }
  }
}

Tip: Use a dedicated venv at C:\Users\YourName\.venvs\mcp-second-brain\ (local SSD) rather than a venv on a network drive. Python loads thousands of small files at startup — on a cloud drive this causes 15–30 s delays and context deadline exceeded errors.

Step 4 — Semantic search (optional, Windows)

Ollama is the easiest path on Windows:

winget install Ollama.Ollama
ollama pull nomic-embed-text

Then add to the env block of your MCP config:

"EMBED_URL": "http://localhost:11434/v1/embeddings",
"EMBED_PORT": "11434"

Ollama starts automatically with Windows. No extra configuration needed.

Alternatively, build llama.cpp for Windows and register it as a scheduled task:

# Register llama-server as a login-triggered scheduled task:
$exe  = "C:\Users\$env:USERNAME\llama.cpp\build\bin\llama-server.exe"
$model = "C:\Users\$env:USERNAME\nomic-embed-text-v1.5.Q8_0.gguf"
$args = "-m `"$model`" --port 11435 --embedding --pooling mean -np 4 -c 2048 --log-disable"

$action   = New-ScheduledTaskAction -Execute $exe -Argument $args
$trigger  = New-ScheduledTaskTrigger -AtLogOn
$settings = New-ScheduledTaskSettingsSet -RestartCount 5 -RestartInterval (New-TimeSpan -Minutes 1)
Register-ScheduledTask -TaskName "llama-embed" -Action $action -Trigger $trigger `
  -Settings $settings -RunLevel Highest -Force

Step 5 — Weekly vault maintenance (optional, Windows)

# Run vault_sleep every Sunday at 02:00
$pythonExe = "C:\Users\$env:USERNAME\.venvs\mcp-second-brain\Scripts\python.exe"
$serverPy  = "C:\path\to\second-brain-mcp\run_sleep.py"
$vaultPath = "C:\Users\$env:USERNAME\second-brain"

$action  = New-ScheduledTaskAction -Execute $pythonExe -Argument "`"$serverPy`"" `
  -WorkingDirectory (Split-Path $serverPy)
$trigger = New-ScheduledTaskTrigger -Weekly -DaysOfWeek Sunday -At 2am
$env_var = [System.Environment]::SetEnvironmentVariable(
  "SECOND_BRAIN_PATH", $vaultPath, "Machine")
Register-ScheduledTask -TaskName "vault-sleep" -Action $action -Trigger $trigger -Force

Development Install (clone)

git clone https://github.com/ddmanyes/second-brain-mcp
cd second-brain-mcp
uv sync
uv run playwright install chromium

Register with Claude Code:

# macOS / Linux
claude mcp add --scope user second-brain \
  --env SECOND_BRAIN_PATH=~/second-brain \
  -- uv run --project /path/to/second-brain-mcp python server.py

# Windows (PowerShell)
claude mcp add --scope user second-brain `
  --env SECOND_BRAIN_PATH="C:\Users\$env:USERNAME\second-brain" `
  -- uv run --project C:\path\to\second-brain-mcp python server.py

Environment Variables

Auto-start (macOS, optional)

# Embedding server — always on, restarts on crash
cp examples/launchd/com.yourname.llama-embed.plist ~/Library/LaunchAgents/
# Edit paths inside the file, then:
launchctl load ~/Library/LaunchAgents/com.yourname.llama-embed.plist

# Weekly vault maintenance — every Sunday 02:00
cp examples/launchd/com.yourname.vault-sleep.plist ~/Library/LaunchAgents/
launchctl load ~/Library/LaunchAgents/com.yourname.vault-sleep.plist

Troubleshooting

Vault Structure

vault/
├── 00-inbox/          # Unprocessed captures — clear daily
├── 10-projects/       # Active projects
├── 20-areas/
│   ├── research/      # Ongoing research domains
│   ├── coding/        # Dev tools and workflows
│   └── consolidated/  # Auto-merged similar notes (Phase 8)
├── 30-resources/      # ← Papers and articles (save_article writes here)
├── 40-archive/        # Compressed originals (auto-managed by vault_sleep)
├── decisions/         # Architecture Decision Records (ADR format)
├── memory/
│   ├── goals.md       # Current priorities — injected at every session start
│   ├── index.md       # Vault map
│   └── rules.md       # Auto-extracted L3 rules — injected at every session start
└── templates/         # Note templates (note, decision, project, research)

Running Tests

uv run pytest tests/ -v
uv run python benchmark.py --quick --markdown   # search latency + accuracy report

References & Acknowledgements

Papers That Directly Inspired This Project

Cognitive Science Foundations

• Ebbinghaus, H. (1885). Über das Gedächtnis. — forgetting curve; basis for access_count / ln(age_days + 1)
• Stickgold, R. (2005). Nature, 437, 1272–1278. — sleep-dependent memory consolidation

Built With

MarkItDown · DuckDB · llama.cpp · nomic-embed-text · FastMCP · Playwright · Anthropic Claude API

Known Issues & Fixes

WAL corruption (Failure while replaying WAL file)

Cause: DuckDB write was interrupted mid-transaction (IDE restart, pkill -9, or machine sleep). Symptom: MCP server crashes on startup; every DB operation fails. Fix:

rm -f ~/.second-brain/vault.db ~/.second-brain/vault.db.wal

Restart the server — it will rebuild a clean DB. Run sync_index to re-index your vault.

Note: If running inside a sandboxed IDE (e.g. Antigravity), the agent cannot delete ~/.second-brain/. Run the command in your local Terminal, or ask Claude Code (VSCode extension) which has full shell access.

vault.db created in the wrong directory

Cause: server.py was launched with a non-home working directory; DuckDB created vault.db relative to cwd. Symptom: ~/.second-brain/vault.db is tiny (< 1 MB) but a large vault.db exists elsewhere. Fix:

# Find the real vault.db
find ~ -name "vault.db" -size +1M 2>/dev/null

# Move it to the correct location
cp /path/to/found/vault.db ~/.second-brain/vault.db

IDE reports no such file or directory for .venv/bin/python

Root cause: The venv was created inside the Google Drive folder. Google Drive sync breaks symlinks — bin/python points to an absolute path on another machine, which doesn't exist locally.

Permanent fix — use a local venv (macOS):

# Create venv on local machine (once per machine)
python3 -m venv ~/.venvs/second-brain
~/.venvs/second-brain/bin/pip install -r /path/to/second-brain/requirements.txt

# Register MCP with Claude Code using the local venv
claude mcp add --scope user second-brain \
  ~/.venvs/second-brain/bin/python \
  /path/to/second-brain/server.py \
  -e PYTHONPATH=/path/to/second-brain \
  -e SECOND_BRAIN_PATH=/path/to/vault

Source code (server.py, etc.) stays on Google Drive and syncs normally. Only the venv lives on the local machine.

Do not create the venv inside the Google Drive folder — it will break on every other machine that syncs it.

Contributing

PRs and Issues welcome. Please open an issue first to discuss significant changes.

License

MIT License — © 2026 Chan Chi Ru. See LICENSE.