projectionlab-mcp
MCP server for ProjectionLab — export, snapshot, and restore your financial plans. Includes a built-in FIRE advisor knowledge base.
README
ProjectionLab MCP Server
MCP server for ProjectionLab — export, snapshot, and restore your financial plans. Includes a built-in FIRE advisor knowledge base for analyzing Financial Independence scenarios.
Why this approach
Most MCP servers that connect to third-party services ask you to paste API keys into config files, environment variables, or .env files. That key then lives on disk, gets passed through the MCP protocol, and travels through the AI conversation — where it can end up in logs, history, or training data.
This server works differently:
- Your API key never leaves the browser. The key is extracted from the ProjectionLab settings page, stored in the browser's
sessionStorage, and used there. The MCP server never sees it. The AI never sees it. It doesn't exist in any config file, environment variable, or conversation log. Other ProjectionLab MCPs store the key in plaintext files on disk or pass it through environment variables. - Everything runs in a browser you can see. This server doesn't make hidden API calls on your behalf. It generates JavaScript that executes in a browser tab where ProjectionLab is open. You can watch every operation happen in real time — open DevTools, inspect network requests, see exactly what's being read or written. Nothing is opaque. Other approaches run headless browsers in the background where you can't observe what's happening.
- Zero infrastructure. No bundled Chromium, no background daemons, no Firefox + GeckoDriver, no Rust toolchain. The only dependencies are the MCP SDK and zod. It works with whatever browser MCP you already have running.
- Snapshots are redacted automatically. When you save financial data locally, any field named
key,apiKey, orpluginKeyis replaced with[REDACTED]before writing to disk. - Published with provenance. Every release is built and published from GitHub Actions via npm trusted publishing. No long-lived tokens, no manual publishes. You can verify that the package on npm was built from this repo —
npm audit signaturesconfirms the cryptographic link between the published tarball and the exact commit that produced it.
The trade-off is that you need a browser automation MCP running alongside this one. But for something as sensitive as your financial data, that transparency is worth it.
Features
- Guided API key setup — securely extracts and validates the key entirely in the browser
- Full data export — returns a script that exports all plans, accounts, income, expenses, and settings
- Local snapshots — save exports as timestamped JSON files with automatic API key redaction
- Snapshot restore — generate scripts to restore plans and current finances from any snapshot
- FIRE knowledge base — curated resource with links to FIRE concepts, withdrawal strategies, and ProjectionLab help docs
Prerequisites
This server generates JavaScript that must be executed in a browser where ProjectionLab is open. You need a browser automation MCP alongside this one:
- Chrome DevTools MCP or Playwright MCP
- A ProjectionLab Premium account with Plugins enabled
API
Tools
-
pl_setup
- Guided two-step API key configuration
- Step 1 (no args): returns the settings URL, a browser extraction script, and setup instructions
- Step 2 (
validationResult): confirms the key is valid and ready - The extraction script reads the key from the DOM, stores it in
sessionStorage, and validates it — all client-side
-
pl_export
- Returns a JavaScript string that exports all ProjectionLab data
- Execute the script in the browser; pass the result to
pl_snapshotto save it - Annotations:
readOnlyHint: true
-
pl_snapshot
- Saves export data as a local JSON file
- Input:
data(required) — the export result from running thepl_exportscript in the browser - Input:
label(optional) — appended to the snapshot filename - API keys are automatically redacted before writing
-
pl_list_snapshots
- Lists all saved snapshots, sorted newest first
- Returns filename, path, creation time, and size
- Annotations:
readOnlyHint: true
-
pl_restore
- Reads a snapshot file and returns JavaScript strings to restore the data
- Input:
snapshotPath(required) — absolute path to the snapshot file - Returns one script per data section (plans, current finances); execute each in order
- Always take a snapshot of the current state before restoring
Resources
- projectionlab://knowledge
- Curated index of ProjectionLab documentation URLs organized by topic
- Covers FIRE concepts, withdrawal strategies, and ProjectionLab features
Usage with Claude Desktop
Add this to your claude_desktop_config.json:
NPX
{
"mcpServers": {
"projectionlab": {
"command": "npx",
"args": ["-y", "@mormubis/projectionlab-mcp"]
}
}
}
Local (from source)
{
"mcpServers": {
"projectionlab": {
"command": "node",
"args": ["/path/to/projectionlab-mcp/dist/index.js"]
}
}
}
Usage with VS Code
Add this to your .vscode/mcp.json:
{
"servers": {
"projectionlab": {
"command": "npx",
"args": ["-y", "@mormubis/projectionlab-mcp"]
}
}
}
Configuration
| Variable | Description | Default |
|---|---|---|
PROJECTIONLAB_SNAPSHOTS_DIR |
Directory for snapshot files | ~/.config/projectionlab/snapshots |
PROJECTIONLAB_BASE_URL |
ProjectionLab base URL | https://app.projectionlab.com/ |
Security
The API key never leaves the browser. During setup, a script runs inside the browser tab to read the key from the DOM, store it in sessionStorage, and validate it against the Plugin API. The only value that returns to the conversation is a boolean.
All other scripts (export, restore) read the key from sessionStorage at runtime. Snapshots are redacted before writing — any field named key, apiKey, or pluginKey is replaced with [REDACTED].
Build
npm install
npm run build
License
This project is licensed under the MIT License. See LICENSE for details.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。