Give ClaudeYour Workflowy.

# create the project
mkdir workflowy-mcp-server && cd workflowy-mcp-server
npm init -y

# folders for tools, services, and compiled output
mkdir -p src/tools src/services

workflowy-mcp-server/
├── package.json
├── tsconfig.json
├── config.json.example     ← template users copy & fill
└── src/
    ├── index.ts            ← entry: MCP server + stdio
    ├── config.ts           ← loads API key + bookmarks
    ├── constants.ts        ← API URLs, defaults
    ├── types.ts            ← shared TypeScript interfaces
    ├── services/
    │   ├── api-client.ts   ← Workflowy API wrapper + rate limit
    │   ├── cache.ts        ← SQLite mirror + FTS5 search
    │   └── sync-engine.ts  ← keeps cache fresh
    └── tools/
        ├── read.ts         ← get node, list children
        ├── search.ts       ← full-text search
        ├── write.ts        ← create / edit nodes
        ├── complete.ts     ← mark done / undone
        ├── bookmarks.ts    ← named shortcuts
        └── sync.ts         ← status + force resync

npm install @modelcontextprotocol/sdk better-sqlite3 bottleneck zod
npm install -D typescript @types/node @types/better-sqlite3

{
  "compilerOptions": {
    "target": "ES2022",
    "module": "Node16",
    "moduleResolution": "Node16",
    "outDir": "./dist",
    "rootDir": "./src",
    "strict": true,
    "esModuleInterop": true,
    "resolveJsonModule": true
  },
  "include": ["src/**/*"]
}

"type": "module",
"scripts": {
  "build": "tsc",
  "start": "node dist/index.js",
  "dev": "tsc --watch"
}

Claude  ⇄ stdio ⇄  MCP server
                      │
          reads/search│          ┌───────────────┐
                      ├─────────▶│ SQLite cache  │  ← instant, no API
                      │          │ (FTS5 search) │
                write │          └───────────────┘
                      ▼                  ▲
              ┌──────────────┐   sync    │
              │ Workflowy API│───────────┘
              └──────────────┘

CREATE TABLE nodes (
  id          TEXT PRIMARY KEY,   -- Workflowy node id
  parent_id   TEXT,               -- for breadcrumbs
  name        TEXT,               -- the node text
  note        TEXT,
  priority    INTEGER,            -- sibling sort order
  layout_mode TEXT,               -- bullets / todo / h1..h3
  completed_at INTEGER,           -- null = active
  modified_at  INTEGER,
  cached_at    INTEGER NOT NULL
);

-- full-text search: this is what makes search instant
CREATE VIRTUAL TABLE nodes_fts USING fts5(
  id UNINDEXED, name, note,
  content='nodes', content_rowid='rowid'
);

import { z } from "zod";

export function registerSearch(server, cache) {
  server.tool(
    "wf_find_node",
    "Search your Workflowy. Use this to LOCATE any node.",
    { query: z.string(), limit: z.number().max(100).default(20) },
    async ({ query, limit }) => {
      // all local SQLite — zero API calls, zero rate limit
      const hits = cache.search(query, limit);
      return { content: [{ type: "text",
        text: formatWithBreadcrumbs(hits) }] };
    }
  );
}

{
  "workflowy_api_key": "YOUR_API_KEY_HERE",
  "bookmarks": {
    "tasks":     { "node_id": "", "label": "Tasks" },
    "capture":   { "node_id": "", "label": "Inbox / Capture" },
    "reference": { "node_id": "", "label": "Reference" }
  },
  "cache_path": "./workflowy-cache.db",
  "sync_interval_minutes": 60
}

cp config.json.example config.json   # then paste your key + ids
npm run build                        # tsc → dist/
node dist/index.js                   # should hang, waiting on stdio = correct

{
  "mcpServers": {
    "workflowy": {
      "command": "node",
      "args": ["/absolute/path/to/workflowy-mcp-server/dist/index.js"],
      "env": { "WORKFLOWY_API_KEY": "your-key-here" }
    }
  }
}

claude mcp add workflowy node /absolute/path/to/dist/index.js \
  --env WORKFLOWY_API_KEY=your-key-here

---
name: workflowy-system
description: >-
  My Workflowy operating system. Trigger whenever the user asks
  to do anything IN Workflowy — search, create, move, read — or
  mentions a column (Tasks, Capture, Reference) or area of their
  outline. Companion to the Workflowy MCP tools: it gives the
  structural map those tools need to operate correctly.
---

# My Workflowy — Structure

## Top-level columns
- **TASKS** — active work. Search here for anything "to do".
- **CAPTURE** — inbox / journal. New raw notes land here.
- **REFERENCE** — the deep archive (most nodes live here).

## Rules the tools must follow
- To LOCATE a node, always use `wf_find_node`. Never dump a tree.
- "Search IN/UNDER <section>" → pass that section as the bookmark
  (WHERE), and the search terms as the query (WHAT).
- Create at the top by default. Never add notes to nodes —
  use child bullets instead.
- Return ALL matching results. Don't cap arbitrarily.

---
name: workflowy-format
description: >-
  Output content as a nested bullet outline ready to paste into
  Workflowy. Trigger on "Workflowy format", "format for Workflowy",
  "WF format", or any list/plan/outline meant for pasting in.
---

# Workflowy Format Rules

- Output ONLY the outline — no preamble, no code fence.
- One item per line. Nest with 2 spaces per level.
- Parent bullets first, children indented beneath.
- No blank lines between bullets (Workflowy reads them as breaks).
- Keep each bullet short; push detail to child bullets, not notes.