Lanes Link
One secure, self-hosted MCP endpoint connecting AI agents to your real accounts (mail, calendar, files, notes), with per-profile separation, explicit permission enforcement, and an append-only audit log.
README
Lanes Link
One secure endpoint between your AI agents and your real accounts, knowledge, and secrets.
Connect your mail, calendar, files, and notes once, and add the memory and skills that only you have. Every agent you use — Claude, ChatGPT, and anything else that speaks MCP — reaches them through a single MCP endpoint that you own and run. Open source, self-hosted, no vendor sitting in the middle of your data.
<picture> <source media="(prefers-color-scheme: dark)" srcset="docs/images/lanes-link-dark.svg"> <img alt="Claude, ChatGPT, and any MCP client reach one Lanes Link endpoint, which you run on your own machine or your own cloud. The profiles it serves sit inside that boundary: Work, holding Gmail, Calendar, and Docs; and Personal, holding Gmail, Memory, and Skills." src="docs/images/lanes-link-light.svg"> </picture>
Why
- Connect once, use everywhere. No per-agent integrations, no re-authorising every new tool.
- You decide what agents can touch. Permissions default to deny, and the runtime enforces them — it does not ask the model to behave.
- Work and personal never mix. Separate profiles, separate credentials, separate stores.
- Every call is recorded. An append-only log of what was reached, and what was refused.
What people use it for
- A personal assistant on your phone. A deployment is a URL, so the claude.ai or ChatGPT app reaches your mail, calendar, and files from anywhere — "what's on tomorrow", "reply to Ana's thread", "find the invoice from March". This is the case that needs your own cloud; everything below works on your machine.
- Inbox triage you sign off on. An agent searches, reads, labels, and composes — and saves a draft for you to send rather than sending it, until you decide otherwise.
- Coding agents that start from your context. Claude Code and Codex reach the same Linear issues, Notion pages, and notes you do, so a session begins from what you already know instead of an empty prompt.
- One set of notes behind every agent. What you have Claude Code write down, claude.ai reads back later — your accumulated context follows you between tools instead of being trapped in whichever one recorded it.
- Your own procedures, followed rather than guessed. How a standup update reads, where an invoice gets filed, who gets cc'd on a contract — written down once, and every agent you use follows the same one.
Quickstart
Needs Bun 1.3.11+. Nothing else — no account anywhere.
$ bun install && bun link # puts `lanes` on your PATH
$ lanes link profile add personal --default
$ lanes link start
ok serving http://127.0.0.1:7337/mcp
profiles: personal
Then, in another shell:
$ lanes link mcp add # every agent installed; or name one: claude, codex
ok registered lanes-link with Claude Code (user scope)
ok registered lanes-link with Codex
Your agents can now use it. Memory, skills, and the vault hold your own material rather than an account, so switching them on costs nothing — one command each, no credentials, no browser. Mail and calendar are the next step. Full quickstart →
What your agent gets
| Manage it with | ||
|---|---|---|
| Connections | your external accounts — mail, calendar, files, issues | lanes link connect |
| Memory | what you want remembered between sessions | lanes link memory |
| Skills | your own procedures, handed to an agent as instructions | lanes link skills |
| Vault | passwords and API keys, released only where you allow it | lanes link vault |
Memory and skills are plain Markdown files, so a text editor and an agent reach the same bytes.
Connect an account
One command per account. Run it again to add a second mailbox, a second calendar, a second anything.
| Connect with | |
|---|---|
| Gmail | lanes link connect gmail |
| Google Drive | lanes link connect drive |
| Google Sheets | lanes link connect sheets |
| Google Docs | lanes link connect docs |
| Google Calendar | lanes link connect calendar |
| Google Tasks | lanes link connect tasks |
| Google Contacts | lanes link connect contacts |
| iCloud Mail | lanes link connect icloud_mail |
| iCloud Calendar | lanes link connect icloud_calendar |
| iCloud Contacts | lanes link connect icloud_contacts |
| iCloud Drive | lanes link connect icloud_drive |
| Notion | lanes link connect notion |
| Linear | lanes link connect linear |
| Gmail (Google MCP) | lanes link connect gmail_mcp |
| Drive (Google MCP) | lanes link connect drive_mcp |
Two things worth knowing up front: lanes link connect icloud sets up Mail, Calendar, and Contacts
together, because one app-specific password covers all three. And Google is the only one that asks
you to register an OAuth client of your own — lanes link connect gmail walks you through it.
Full guide — what each one gives your agent, what it needs, and adding your own: docs/connect.md.
Run it anywhere
The same code, the same config, in all three. Only the storage adapters change.
| Local | Your own cloud | Lanes Cloud | |
|---|---|---|---|
| Runs on | your machine | your GCP project, on Cloud Run | managed for you |
| Needs | Bun, nothing else | a Google Cloud billing account | — |
| Set up with | lanes link start |
lanes link deploy |
join the waitlist |
| Reachable from | that machine | anywhere, including your phone | anywhere |
| Status | ready | ready | coming soon |
Local is the fastest way to start, and where most people stay. Your own cloud is what you
want if you need to reach it from claude.ai, ChatGPT, or a phone — lanes link deploy creates the
project, the bucket, the service account, and the revision on its first run. Lanes Cloud is the
managed version; because it is the same data model, a workspace you build today moves across rather
than being rebuilt.
Join the waitlist to hear when it opens.
Docs
- Quickstart — from nothing to a working endpoint
- Connect your accounts — every provider, and what each one needs
- Add it to your agent — Claude Code, Codex, Claude Desktop, claude.ai, ChatGPT
- Deploy to your own cloud — five commands to a URL
- Full reference — architecture, configuration, the CLI, the security model, writing a provider, and the decision records
Security
Lanes Link holds live credentials to your email and documents. The
security model states its limits plainly rather than implying
guarantees the code does not deliver — read it before you trust it with an account. To report a
vulnerability, see SECURITY.md; please do not open a public issue.
License
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。