obsidian-mcp-secure

obsidian-mcp-secure

Secure Model Context Protocol server that turns your Obsidian vault into a reliable data source for any MCP-compatible AI client — built from scratch with OWASP Top 10 controls and full audit logging.

Listed on the official Anthropic MCP Registry as io.github.dewtech-technologies/obsidian-mcp-secure.

---

🧭 Positioning — this is NOT a plugin for Obsidian

It's the opposite: it's a bridge that lets Claude Desktop (or any MCP client) read and write inside Obsidian safely. Your AI assistant stays where it lives; your vault becomes a structured, auditable datasource it can reach.

┌─────────────────┐   MCP    ┌──────────────────────┐   HTTP   ┌────────────────────┐   FS   ┌─────────────┐
│                 │  stdio   │                      │  :27123  │                    │        │             │
│ Claude Desktop  │ ───────▶ │ obsidian-mcp-secure  │ ───────▶ │  Local REST API    │ ─────▶ │  Vault .md  │
│  (AI client)    │          │  (this package)      │          │ (Obsidian plugin)  │        │             │
└─────────────────┘          └──────────────────────┘          └────────────────────┘        └─────────────┘

Role in the pipeline	Component
Where you talk	Claude Desktop (or any MCP client)
Bridge / access control	obsidian-mcp-secure (this package)
Data gateway inside Obsidian	Local REST API plugin (by Adam Coddington)
Your knowledge	.md files in your vault

One-liner: Claude is the brain, this MCP is the arm, Obsidian is the memory.

Why another Obsidian + AI integration?

There are plugins that put Claude inside Obsidian. This is the inverse, and it exists because:

• Your assistant is Claude Desktop — that's where the general-purpose conversations happen. Your notes become one of many contexts Claude can reach, alongside web, GitHub, filesystems, etc.
• Security is a first-class concern — deliberate attack surface, no shell access, path traversal blocked, inputs validated with Zod, every call audited.
• Zero build, zero account — npx obsidian-mcp-secure and done. Works on Windows, macOS, Linux the same way.
• Composability — combine this MCP with fetch, filesystem, git, GitHub, etc., and Claude can cross-reference your vault with external sources in a single conversation.

---

🛠️ Available Tools

Tool	Purpose
read_note	Read a note by path
list_notes	List files/folders in the vault or a subdirectory
create_note	Create a new .md note
edit_note	Overwrite an existing note (previous content goes to the audit log)
delete_note	Delete a note — requires confirm: true (Zod rejects otherwise)
search_notes	Full-text / tag search using Obsidian's own search engine
find_note_by_name	Find notes by partial name — case-insensitive, no exact path needed
list_tags	Enumerate all tags in the vault with usage count; sortable by name or frequency
create_backlinks	Add [[wikilinks]] to a ## Relacionadas section in a note — explicit and auditable

---

🔒 Security — OWASP Top 10

Control	Implementation
A01 — Broken Access Control	Path traversal blocked (../, ..\\, encoded variants); .md extension enforced
A02 — Cryptographic Failures	API key read from .env or process env; never hardcoded, never logged
A03 — Injection	All inputs validated with Zod schemas; no eval, no exec, no shell
A04 — Insecure Design	512 KB max note size; 50-result cap on search; destructive ops require explicit confirm: true
A05 — Security Misconfiguration	Only 127.0.0.1 / localhost accepted as host
A09 — Logging & Monitoring	Full audit log via winston with size-based rotation (5 MB / 10 files)

Every tool call emits an audit line with action, params (sanitized), success, error, and timestamp.

---

⚡ Installation

Prerequisites

1. Obsidian Desktop with a vault open
2. The Local REST API plugin (by Adam Coddington) — install from Community Plugins, enable it, and:
   • Turn on "Enable Non-encrypted (HTTP) Server" (simpler than HTTPS self-signed certs)
   • Copy the API Key shown in the plugin settings
3. Node.js 18+
4. Claude Desktop (or another MCP-compatible client)

Configure Claude Desktop

Open %APPDATA%\Claude\claude_desktop_config.json on Windows (or ~/Library/Application Support/Claude/claude_desktop_config.json on macOS) and add:

{
  "mcpServers": {
    "obsidian-secure": {
      "command": "npx",
      "args": ["-y", "obsidian-mcp-secure"],
      "env": {
        "OBSIDIAN_API_KEY": "your-api-key-from-the-plugin",
        "OBSIDIAN_HOST": "http://127.0.0.1",
        "OBSIDIAN_PORT": "27123",
        "LOG_DIR": "C:/path/to/your/logs"
      }
    }
  }
}

Windows tip: if npx fails silently, switch "command": "npx" to "command": "npx.cmd". Some Claude Desktop builds don't resolve bare npx on PATH.

Restart Claude Desktop (tray → Quit, then reopen) and the 9 tools will show up under obsidian-secure.

---

🤝 Recommended companions

The real power of MCPs is composability. To reproduce the "read my note → fetch a URL → tell me if I'm applying it correctly" workflow, add the official fetch MCP alongside this one:

{
  "mcpServers": {
    "obsidian-secure": { "...": "as above" },
    "fetch": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-fetch"]
    }
  }
}

Now Claude has both your vault and the live web in a single conversation.

---

💬 Example prompts

With obsidian-secure + fetch enabled:

"Read my note Projeto API Atendimento.md, then fetch https://developers.facebook.com/docs/whatsapp and tell me if my implementation matches the latest best practices."

"Search my vault for the tag #ideia and summarize the three ideas that appear most often. Then create a new note called Ideias recorrentes.md with the summary."

"Read Atomic Habits - Resumo.md, fetch https://jamesclear.com/atomic-habits, and point out where my notes drifted from the original."

Claude will orchestrate the tool calls automatically — no manual chaining.

---

🧩 Comparison with in-Obsidian plugins

If your workflow lives inside Obsidian's sidebar, plugins like obsidian-claude-code are the right fit. This MCP targets a different shape:

Dimension	obsidian-claude-code (in-Obsidian)	obsidian-mcp-secure (this)
Where the AI lives	Sidebar inside Obsidian	Claude Desktop (or any MCP client)
Setup	git clone + bun build	npx obsidian-mcp-secure
Tools	Read/Write/Edit + Bash + Grep + Glob + WebFetch	9 purpose-built, Zod-validated tools
Security posture	Full shell access to dev machine	Tight allowlist, audited, OWASP Top 10
Distribution	Manual clone, requires Bun	npm + official MCP Registry
Composability with other sources	Inside its own sandbox	Any MCP-compatible client can mix it with fetch, GitHub, filesystem, etc.
Best for	Dev who lives in Obsidian	Professional whose main surface is Claude Desktop

Both are valid — they occupy different niches.

---

🔧 Environment variables

Variable	Required	Default	Description
OBSIDIAN_API_KEY	✅	—	API key from the Local REST API plugin
OBSIDIAN_HOST		http://127.0.0.1	Host (only 127.0.0.1 and localhost are accepted)
OBSIDIAN_PORT		27123	Port of the plugin's HTTP server
LOG_DIR		./logs	Directory for the audit log files

---

🗺️ Roadmap

✅ Shipped in v1.2.1

• Bug fix: find_note_by_name searches full path (folder + filename)
• Bug fix: list_tags normalizes all API response formats (object, array of strings, array of objects with tagCount/taggedFilesCount)

✅ Shipped in v1.2.0

• DXT package for one-click install in Claude Desktop (npm run build:dxt)

✅ Shipped in v1.1.0

• find_note_by_name — partial, case-insensitive name match across the entire vault
• create_backlinks — connect related notes with [[wikilinks]] (explicit, auditable)
• list_tags — enumerate all tags in the vault with usage count
• Unit test suite (70 tests — utils, handlers, HTTP client) with Vitest
• CI pipeline on every PR: tests + coverage + npm audit + static security analysis

🔜 Up next

• Smithery listing
• Read-only mode flag for shared / multi-user setups

Ideas and PRs welcome — see CONTRIBUTING.md.

---

📜 License

MIT — see LICENSE.

🙏 Credits

• Model Context Protocol by Anthropic
• Local REST API plugin by Adam Coddington — the foundation that makes this possible
• Built at Dewtech by Wanderson Leandro

---

Security issues? See SECURITY.md for disclosure instructions.