microsoft-todo-mcp
A remote MCP server for Microsoft To Do that enables read/write access to tasks, lists, and checklists from ChatGPT or Claude, with multi-account support and Azure deployment.
README
microsoft-todo-mcp
A remote Model Context Protocol server
for Microsoft To Do, deployable on Azure and usable as a custom connector
in both ChatGPT and Claude. Multi-account (personal + work
Microsoft accounts), multi-profile (/mcp/personal, /mcp/work, or any
custom alias), single-owner administration.
Architecture at a glance
flowchart TB
subgraph Clients
ChatGPT[ChatGPT custom connector]
Claude[Claude custom connector]
Browser[Owner's browser -> /accounts]
end
App["microsoft-todo-mcp\n(Azure Container Apps, scale-to-zero)"]
KV[(Key Vault)]
ST[(Storage: Table + Blob)]
Graph[[Microsoft Graph]]
ChatGPT -- Bearer JWT --> App
Claude -- Bearer JWT --> App
Browser --> App
App --> KV
App --> ST
App -- per-connection token --> Graph
Full details: docs/architecture.md.
Why this exists
Microsoft To Do has no first-party remote MCP connector. This project lets you talk to your Microsoft To Do lists from ChatGPT or Claude, across however many personal/work Microsoft accounts you have, from a small self-hosted deployment you fully control (your own Azure subscription, your own OAuth authorization server, your own encrypted token storage).
Feature summary
- MCP tools: full read/write coverage of task lists, tasks, and
checklist items, plus account/profile discovery tools — see the tool
list in
src/tools/. - Multi-account routing that never guesses which Microsoft account a
tool call should hit — see
docs/multi-account.md. - Two independent OAuth systems, documented in
docs/oauth.md: this server as an OAuth 2.1 authorization server toward ChatGPT/Claude, and this server as an OAuth client toward Microsoft Entra ID/Graph. - Encrypted token storage: AES-256-GCM MSAL token caches in Blob Storage, hashed OAuth codes/refresh tokens in Table Storage, secrets in Key Vault, all accessed via a least-privilege user-assigned managed identity — no client secrets or connection strings in the running production process.
- Owner-only admin UI at
/accountsfor connecting accounts, binding profiles, reauthorizing, revoking, and deleting connections. - Azure Container Apps, consumption plan, scale-to-zero — near-zero
idle cost; see the cost estimate in
docs/azure-deployment.md.
Quick start (local development)
npm install
cp .env.example .env # fill in real values — see docs/azure-deployment.md
npm run dev
npm run dev starts the server with tsx watch. You'll need a reachable
Storage account (Azurite works for local dev) and Key Vault (or adapt
src/config/secrets.ts for local secret sources) before OAuth flows will
actually work end-to-end — see docs/operations.md.
Deploying to Azure
./scripts/login.ps1
./scripts/validate-environment.ps1
$region = ./scripts/select-region.ps1
./scripts/provision.ps1 -ResourceGroupName rg-microsoft-todo-mcp -Location $region -ResourceSuffix prod01 `
-AzureClientId <graph-app-client-id> -AzureTenantId <tenant-id> `
-PublicBaseUrl https://todo.h-aa.dk -BudgetNotificationEmail you@example.com
./scripts/deploy.ps1 -ResourceGroupName rg-microsoft-todo-mcp -ContainerAppName todomcp-app-prod01 `
-Image ghcr.io/<you>/microsoft-todo-mcp:sha-abc123 -Build
./scripts/configure-domain.ps1 -ResourceGroupName rg-microsoft-todo-mcp `
-ContainerAppEnvironmentName todomcp-env-prod01 -ContainerAppName todomcp-app-prod01 `
-DomainName todo.h-aa.dk -Location $region
Full walkthrough: docs/azure-deployment.md.
DNS specifics: docs/dns.md.
Connecting ChatGPT / Claude
- Connect at least one Microsoft account and bind a profile from
/accounts—docs/multi-account.md. - Add a custom connector pointing at
/mcp/personal,/mcp/work, or/mcp(universal) —docs/chatgpt-setup.md/docs/claude-setup.md.
Security model
Summarized in docs/security.md; vulnerability
reporting in SECURITY.md.
Cost estimate
Roughly 10–25 DKK/month depending on whether Log Analytics is enabled
— see the itemized breakdown in
docs/azure-deployment.md.
Teardown
az group delete --name rg-microsoft-todo-mcp --yes --no-wait
Project layout
src/ Server source (mcp, graph, oauth, storage, crypto, tools, admin, http)
infra/ Bicep: Container Apps (consumption, scale-to-zero), Key Vault, Storage, RBAC, budget
scripts/ PowerShell 7 provisioning/deploy/ops scripts
docs/ Architecture, deployment, DNS, OAuth, security, connector setup, operations
.github/ CI, CodeQL, secret scanning, OIDC-based deploy workflow
tests/ Vitest unit tests + Playwright OAuth flow tests
License
MIT.
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。