devonthink-mcp-plus

Supplementary MCP server for DEVONthink, designed to work alongside the stock mcp-server-devonthink.

The stock MCP handles write operations (create, update, delete, tag, classify). This server fills the read gap: returning document content alongside search results, paginating large files, and loading entire project folders as context — all without extra round-trips.

Why this exists

The stock mcp-server-devonthink search returns only metadata (name, path, score). To actually read a document you need a second tool call. For large PDFs or emails the full content blows out the context window with no way to paginate. This server fixes both problems by surfacing DEVONthink's existing full-text index directly.

Tools

dt_search_with_excerpts

Search DEVONthink and get plain-text excerpts inline — one call, no follow-up needed.

Parameter	Type	Default	Description
query	string	required	Search query. Supports DEVONthink operators: name:, tag:, kind:, etc.
database	string	all	Restrict to a specific database name
limit	number	10	Max results (capped at 50)
excerpt_chars	number	600	Plain-text characters to include per result

Supported formats: Markdown, PDF, Email (.eml), Excel (.xlsx), Word (.docx) — anything DEVONthink has indexed.

---

dt_get_content_chunked

Read a document in pages. Use offset + limit to walk through large files without filling the context window. The response includes total_chars, has_more, and next_offset so you know whether to keep reading.

Parameter	Type	Default	Description
uuid	string	required	Record UUID (from search results)
offset	number	0	Character position to start from
limit	number	4000	Max characters to return

---

dt_get_group_context

Load a plain-text snapshot of every document inside a DEVONthink group. Useful for priming project context before starting a task.

Parameter	Type	Default	Description
group_path	string	required	Database-relative path, e.g. /ONVIF_Offering/Longse
database	string	required	Database name, e.g. Active_Work
max_chars_per_doc	number	800	Max plain-text chars per document
max_docs	number	20	Max documents (direct children only, no sub-groups)

---

Requirements

• macOS (uses osascript to talk to DEVONthink)
• DEVONthink 3 or later, running
• Node.js 18+

Installation

git clone https://github.com/danielkuang666/devonthink-mcp-plus.git
cd devonthink-mcp-plus
npm install

Register with Claude Code:

claude mcp add --scope user devonthink-plus node /path/to/devonthink-mcp-plus/server.js

Verify it's connected:

claude mcp list

Usage with Claude Code

# Search with content
dt_search_with_excerpts(query="ONVIF ProfileM", database="Active_Work")

# Read a large PDF in pages
dt_get_content_chunked(uuid="CA5B35C5-...", offset=0, limit=4000)
dt_get_content_chunked(uuid="CA5B35C5-...", offset=4000, limit=4000)  # next page

# Load a project folder as context
dt_get_group_context(group_path="/ONVIF_Offering/Longse", database="Active_Work")

How it works

• JXA (JavaScript for Automation): used for search and content retrieval, returns structured JSON
• AppleScript: used for group navigation (get record at path), which is more reliable than JXA for path-based lookups
• Scripts are written to temp files before execution to avoid shell-escaping issues
• DEVONthink's plainText property provides already-extracted text for all indexed formats — no separate OCR or parsing needed

Pair with the stock MCP

This server is read-only. For write operations, keep mcp-server-devonthink installed alongside it:

Need	Use
Search + read content	devonthink-plus (this server)
Create / update / delete records	mcp-server-devonthink
Add tags, set properties	mcp-server-devonthink
Classify, compare, AI summarize	mcp-server-devonthink

Versioning

# After making changes
git add server.js
git commit -m "feat: description"
git push
git tag v0.x.0
git push origin v0.x.0
gh release create v0.x.0 --generate-notes