second-brain MCP — Claude Code MCP server: a knowledge vault
A local Markdown vault that your AI agent reads, writes and tidies up on its own. Useful when notes and papers pile up faster than you can file them.
claude mcp add mcp-second-brain -- uvx mcp-second-brainWhat second-brain MCP does
second-brain MCP exposes a folder of plain Markdown files to your agent over MCP. One call to save_article pulls in a URL or PDF, turns it into Markdown, OCRs the figures, embeds the result and links it to related notes. Search covers both note text and figure content, so search_figures can return the exact panel you remember from somewhere in the library.
The vault also thins itself out. Notes nobody opens get compressed over time, and ranking follows the Ebbinghaus forgetting curve. At session start get_context reloads goals, rules and the top-ranked notes, so the agent does not begin from nothing. The index can be rebuilt whenever you like with sync_index; the files stay the source of truth.
You need Python 3.11–3.13 and Chromium for Playwright. The vault directory and templates appear on first run. DuckDB is the default index and works offline; Postgres with pgvector is there for setups across several machines. Installing from PyPI may give you an older build than the source tree, so take the newest version from the repository if that matters.
What it helps with
- save a paper or PDF into a knowledge base with one call
- find a specific figure by description across the whole library
- link a new note to ones you already have
- compress stale notes so recall stays cheap
- reload goals and rules when a session starts
Example prompts for the agent
- Save this article to the vault with
save_article - Find the UMAP melanocyte panel using
search_figures - Load my session context: goals, rules and top notes
How to install second-brain MCP
Platform CLI
claude mcp add mcp-second-brain -- uvx mcp-second-brainuvx runs the command named after the package; if the README names another one, use uvx --from mcp-second-brain <command>
Platform CLI
codex mcp add second-brain-mcp -- uvx mcp-second-brainuvx runs the command named after the package; if the README names another one, use uvx --from mcp-second-brain <command>
Config file
{
"mcpServers": {
"second-brain-mcp": {
"command": "uvx",
"args": [
"mcp-second-brain"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mcp-second-brain <command>
Config file
{
"servers": {
"second-brain-mcp": {
"type": "stdio",
"command": "uvx",
"args": [
"mcp-second-brain"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from mcp-second-brain <command>
Platform CLI
gemini mcp add second-brain-mcp uvx mcp-second-brainuvx runs the command named after the package; if the README names another one, use uvx --from mcp-second-brain <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about second-brain MCP
How do I install second-brain MCP in Claude Code?
Install the package and the browser first: pip install mcp-second-brain and playwright install chromium. Then register the server with claude mcp add mcp-second-brain -- uvx mcp-second-brain.
What does second-brain MCP do?
It keeps a personal Markdown vault: saves articles and PDFs, OCRs figures, builds embeddings, auto-links notes and compresses the ones you stopped reading.
What does the server need to run?
Python 3.11–3.13 and Chromium for Playwright. The index defaults to DuckDB and runs offline; Postgres with pgvector is available for multi-machine setups.
