Mythos

Obsidian MCP + Claude Code is an integration pattern that connects an 📝Obsidian vault to 📝Claude Code via a Model Context Protocol server, giving Claude persistent read/write access to a user's notes during coding sessions.

The pattern has become one of the most popular approaches to giving Claude long-term memory. Several open-source MCP servers expose vault contents as tool calls — read_note, search_vault, create_note — so Claude can retrieve context mid-session without manual copy-pasting. The setup appeals to developers and researchers already living in Obsidian who want their AI assistant to remember prior work, preferences, and project context. As vaults scale, the pattern encounters structural limitations around retrieval quality, access control, and mobile support that purpose-built systems like @MythOS were designed to address.

Key Facts

  • Category: AI memory integration pattern
  • Components: Obsidian vault + MCP server + Claude Code MCP client
  • Popular servers: obsidian-mcp (direct filesystem), mcp-obsidian (REST API plugin), obsidian-mcp-tools (semantic search + Templater), obsidian-mcp-server (structured knowledge management)
  • Transport: stdio (local process only — no remote or mobile access)
  • Access model: Full read/write/delete to the configured vault path; no per-note permissions
  • Prerequisite: Claude Code desktop; some servers also require Obsidian to be running

How It Works

MCP server configuration. A server is added to claude_desktop_config.json or .mcp.json with the vault's absolute path. The simplest option, obsidian-mcp, installs via npx -y @smithery/cli install obsidian-mcp --client claude. The REST API variant, mcp-obsidian, requires Obsidian's Local REST API community plugin running first, configured with OBSIDIAN_API_KEY, OBSIDIAN_HOST, and OBSIDIAN_PORT environment variables.

CLAUDE.md as baseline context. A CLAUDE.md file placed in the vault root is read automatically by Claude Code at session start. It typically contains user preferences, current projects, and communication style. The MCP server supplements this with on-demand access to deeper vault content.

Session logging via daily notes. Claude reads today's daily note at session start, works with the user, then writes a summary at session end. Over time, daily notes become a rolling memory log — a pattern that works well at small scale but produces unstructured accumulation over time.

On-demand retrieval. During coding sessions, Claude searches the vault for relevant notes — meeting transcripts, design decisions, past debugging sessions — using the MCP server's search tools rather than requiring manual lookup.

Why It Matters

The Obsidian MCP pattern represents the first widely-adopted approach to giving Claude persistent, user-controlled memory outside of Anthropic's built-in features. It demonstrated real demand for AI memory systems and validated the MCP protocol as a viable bridge between knowledge tools and AI agents.

The pattern also exposed the structural ceiling of adapting note-taking applications for AI retrieval. Most Obsidian MCP servers load entire files into the context window rather than performing semantic search, which degrades at scale — hundreds of notes means token waste on irrelevant content. The vault is exposed without per-note permissions, so personal journals and sensitive client notes are equally accessible to any query. The stdio transport restricts usage to Claude Code on desktop, with no path to mobile or web access. And Claude's write access introduces overwrite risk with no separation between human-authored and AI-authored content.

These limitations map directly to the design decisions in purpose-built alternatives. @MythOS uses vector embeddings for semantic retrieval, three-tier visibility with audience tags for access control, HTTP transport with OAuth 2.0 for cross-platform access, and a dedicated notes field that separates AI contributions from user content.

FAQ

What Obsidian MCP servers are available for Claude Code?

The four main options are obsidian-mcp (direct filesystem access, simplest setup), mcp-obsidian (uses Obsidian's Local REST API plugin), obsidian-mcp-tools by jacksteamdev (adds semantic search and Templater integration), and obsidian-mcp-server by cyanheads (comprehensive knowledge management operations). Each takes a different approach to vault access.

Does Obsidian need to be running for the MCP server to work?

It depends on the server. obsidian-mcp accesses vault files directly through the filesystem — Obsidian does not need to be open. mcp-obsidian communicates through the Local REST API plugin, which requires Obsidian to be running.

Can Obsidian MCP work on mobile or Claude's web interface?

No. The MCP integration uses stdio transport, which requires a local process on the same machine as Claude Code. There is no built-in way to bridge it to Claude's web or mobile interfaces without running custom server infrastructure.

Is the Obsidian MCP setup secure?

The MCP server has full access to whatever vault path is configured. There is no sandboxing, no per-note permissions, and no audit log. The community recommendation is to maintain backups before enabling write access — particularly for vaults containing sensitive or client-related content.

Can Obsidian and MythOS be used with Claude Code simultaneously?

Yes. Claude Code supports multiple MCP servers in the same session. Users can keep Obsidian for local note-taking while using MythOS for AI memory and semantic retrieval. This is a common configuration during transition.

Related

Contexts

Created with 💜 by One Inc | Copyright 2026