Obsidian + Claude Code: Q2 2026 Update

This Is a Quarterly Update

Between January and May 2026, the Obsidian + AI ecosystem shifted significantly. The original series covered vault foundations, MCP setup, AI-powered skills, life tracking, and professional workflows across five articles. Four months later, three of those articles need corrections.

This is the first quarterly update — a single reference point covering what changed, what to do about it, and what’s coming next. Expect the Q3 update in September 2026.

The short version: Setup is now simpler (built-in MCP, no external server), Claude Code is significantly more capable (Opus 4.7, auto memory, plugins), and skill syntax has breaking changes you need to fix.

What’s Still True

Not everything changed. These foundations remain solid:

• PARA method for vault organization — still the recommended approach (Part 1)
• Template system — daily notes, meetings, project kickoffs all work identically
• Frontmatter standards — same fields, same conventions
• Daily life tracking — car maintenance, reading logs, journaling (Part 3)
• The philosophy — vault as source of truth, AI as assistant not replacement

If you only read Parts 1 and 3, your workflow is unaffected. The changes hit the AI integration layer — how Claude Code connects to and interacts with your vault.

The Big Infrastructure Shift

Three changes converged in early 2026 to fundamentally simplify the Obsidian + AI stack:

1. Built-in MCP Server (Local REST API v4.0.0)

The Local REST API plugin now ships its own MCP server. No external Python process, no uvx, no separate package to maintain. The plugin is the server.

• Endpoint: https://127.0.0.1:27124/mcp/ (HTTPS) or http://127.0.0.1:27123/mcp/ (HTTP)
• Transport: Streamable HTTP with Bearer token auth
• Tools: 16 built-in (vault CRUD, search, tags, commands, periodic notes)
• Extensible: Third-party Obsidian plugins can register additional MCP tools

Breaking changes from v3.x: The Apply-If-Content-Preexists header was renamed to Reject-If-Content-Preexists, and Dataview DQL support was removed (replaced by JsonLogic queries).

2. Obsidian CLI (v1.12.0+)

Obsidian now has an official CLI explicitly designed for “agentic tools.” It bypasses the REST API entirely — interacting with vaults through subprocess calls that respect Obsidian’s indexing and sync.

obsidian read vault="Main-vault" path="folder/note.md"
obsidian search vault="Main-vault" query="search term"
obsidian daily:append vault="Main-vault" content="New entry"
obsidian create vault="Main-vault" path="folder/new.md" content="..."
obsidian tags vault="Main-vault"

This is particularly useful for skills that need to append to daily notes or create new files — it integrates with Obsidian Sync automatically.

3. Obsidian Sync Headless

Announced February 2026, this runs Obsidian Sync without a GUI. It enables server-side automation and gives agentic tools vault access without full computer access. Useful for CI pipelines, scheduled tasks, and remote agent workflows.

Revised Quick Start: Full Setup

This replaces the original Quick Start guide. Total time: ~5 minutes (down from 15).

Prerequisites

• Obsidian v1.12.0+ (check: Help → About)
• Claude Code (install: claude install or via code.claude.com)
• Local REST API plugin v4.0.0+ (the MCP server)

Step 1: Install the Local REST API Plugin

1. Open Obsidian → Settings → Community Plugins → Browse
2. Search for “Local REST API”
3. Install and enable it
4. If already installed, check the version is 4.0.0 or higher (Settings → Community Plugins → Local REST API → check version). If you don’t see a version number displayed, assume it’s an older version — uninstall the plugin and reinstall it fresh from the community browser. See the official install instructions if you run into issues.

Cleanup: With v4.0.0’s built-in MCP server, you no longer need separate MCP bridge plugins. If you have Obsidian MCP Tools or Semantic Notes MCP installed, you can safely remove them — the Local REST API plugin now handles everything those provided. Removing them reduces plugin conflicts and simplifies your stack.

Step 2: Get Your API Key

1. Go to Settings → Local REST API
2. Copy your API key (auto-generated on first enable)
3. Note the HTTPS port (default: 27124)
4. Ensure “Enable HTTPS” is checked (recommended for security)

Step 3: Add to Claude Code

claude mcp add --transport http --scope local \
  --header "Authorization: Bearer YOUR_API_KEY_HERE" \
  obsidian-vault https://127.0.0.1:27124/mcp/

Replace YOUR_API_KEY_HERE with the key from Step 2.

What the flags mean:

• --transport http — uses Streamable HTTP (the current standard; SSE is deprecated)
• --scope local — available only in this project directory (use --scope user for global access)
• --header — passes your API key for authentication
• obsidian-vault — the name you’ll see in Claude Code
• The URL — your plugin’s built-in MCP endpoint

Step 4: Verify the Connection

Start Claude Code and ask:

“List the tools available from the obsidian-vault MCP server”

You should see 16 tools including vault file operations, search, tags, and periodic notes. Then test with:

“Search my Obsidian vault for notes about [something you know exists]”

Alternative: Manual JSON Configuration

If you prefer to edit the config file directly:

{
  "mcpServers": {
    "obsidian-vault": {
      "transport": "http",
      "url": "https://127.0.0.1:27124/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY_HERE"
      }
    }
  }
}

Add this to ~/.claude/settings.json under the mcpServers key.

Alternative: Obsidian CLI Approach

If you prefer the CLI (no plugin required for basic file operations):

claude mcp add --transport stdio --scope local \
  obsidian-cli -- obsidian mcp serve --vault "Main-vault"

This uses Obsidian’s native CLI as an MCP server. Simpler, but fewer tools (no search, no commands). Best combined with the built-in server for a complete setup.

Troubleshooting

Tip: If you have many MCP servers, Claude Code defers tool loading to save context. Add "alwaysLoad": true to your obsidian-vault server config to ensure its tools are always available:

{
  "mcpServers": {
    "obsidian-vault": {
      "transport": "http",
      "url": "https://127.0.0.1:27124/mcp/",
      "headers": { "Authorization": "Bearer YOUR_KEY" },
      "alwaysLoad": true
    }
  }
}

Claude Code: What Changed

Between January and May 2026, Claude Code went through ~140 releases. Here’s what matters for Obsidian workflows:

Key Implications for Your Workflow

Auto memory means Claude Code remembers your vault structure, naming conventions, and preferences across sessions. You no longer need to re-explain your PARA setup each time.

Opus 4.7 significantly improves synthesis quality — cross-note connections, pattern recognition, and report generation are noticeably better than January.

The plugin system means skills can now be packaged and shared. The Obsidian community is building plugin-bundled skills you can install with one command.

/goal command lets you set a completion condition and Claude works across multiple turns autonomously. Example: /goal "Generate my weekly summary and save it to the vault" — it will read notes, synthesize, write the file, and verify without intervention.

Updated Skills Workflow

If you followed Part 2 and created custom skills, you need to update them.

What Changed

1. Skills and commands merged — .claude/skills/ and .claude/commands/ now work identically. Skills are recommended.
2. Argument syntax — $ARGUMENTS.0 is now $ARGUMENTS[0]. Shorthand: $0, $1.
3. New frontmatter fields — effort, model, paths, hooks, when_to_use, argument-hint.
4. Type-to-filter — the /skills menu now has a search box and can sort by token count.
5. Description cap raised — from 250 to 1,536 characters for better discoverability.

Migration Example: Status Report Skill

Before (January 2026):

---
name: status-report
description: Generate a weekly status report from recent daily notes
---

Read the last 7 days of daily notes from $ARGUMENTS.0 (default: "Daily Notes")
using batch_get_file_contents...

After (May 2026):

---
name: status-report
description: Generate a weekly status report from recent daily notes. Reads the last 7 days, extracts completed tasks and project updates, and formats them into a professional report using your template.
argument-hint: "[folder-path]"
effort: high
when_to_use: "When the user asks for a status report, weekly summary, or standup prep"
---

Read the last 7 days of daily notes from $ARGUMENTS[0] (default: "Daily Notes")
using the obsidian-vault MCP server's search and file operations...

New Frontmatter Fields Worth Using

The MCP Ecosystem in May 2026

The built-in server covers most use cases, but the ecosystem has grown significantly. Here’s when to look beyond it:

When to Upgrade

• Stick with built-in if you have <1000 notes and basic search/CRUD is sufficient
• Add enquire-mcp if you’re asking “find notes related to X” and keyword search isn’t cutting it
• Add kObsidian if you use Tasks plugin, Kanban boards, or Canvas and want AI access to those
• Add MegaMem if you want to query “what changed about Project X between March and April” or track entity relationships

You can run multiple MCP servers simultaneously. The built-in handles CRUD while a specialized server handles advanced search.

Roadmap: What to Watch Before Q3

Here’s what’s practical now, what’s emerging, and what to monitor for the September update.

Stable Now — Migrate Today

• Built-in MCP server — production-ready, replaces the external Python setup
• Obsidian CLI — stable for file operations, daily note appending
• Updated skill syntax — $ARGUMENTS[0], new frontmatter fields
• Streamable HTTP transport — the standard going forward

Emerging — Worth Experimenting With

• Plugin extension API — third-party Obsidian plugins can register MCP tools with the built-in server. Watch for community plugins that add specialized tools (e.g., a Dataview plugin registering query tools)
• enquire-mcp’s GraphRAG — graph-boosted retrieval is promising for large vaults but still maturing
• Obsidian Sync Headless — enables automation without the app running, but documentation is sparse
• Claude Code plugins with bundled skills — the packaging story is solidifying; expect installable skill packs for common Obsidian workflows

Wait and See — Q3 Will Clarify

• MegaMem’s knowledge graph approach — powerful concept but requires Neo4j, heavy setup
• kObsidian’s LLM-Wiki pattern — interesting but the 66-tool surface area is overwhelming
• Obsidian’s official AI stance — they’re enabling the ecosystem (CLI, Sync Headless) rather than building their own AI features. Whether that changes will shape the roadmap significantly

What We’ll Cover in Q3

• Performance benchmarks across MCP servers
• Plugin extension API maturity and community adoption
• Any new Obsidian or Claude Code releases that shift the landscape
• Reader feedback on this setup working in practice

Migration Checklist

If you followed the original series, here’s what to update:

• Update Local REST API plugin to v4.0.0+ (Settings → Community Plugins → Check for updates)
• Remove old MCP server if using external Python/uvx approach: claude mcp remove obsidian-mcp
• Add built-in MCP server using the Step 3 command above
• Update skill syntax — find-replace $ARGUMENTS.0 → $ARGUMENTS[0] in all ~/.claude/skills/ files
• Update MCP scope references — project → local, global → user in any documentation or scripts
• Add when_to_use to skill frontmatter — improves auto-suggestion
• Test — run your most-used skill and verify it works end-to-end
• Optional: Add alwaysLoad: true to your MCP config if tools aren’t appearing

Series Index

This is the Q2 2026 update. Next update: September 2026. Found an error or have a suggestion? Open an issue or reach out.