MCP Server for Microsoft Planner

An unofficial MCP server that connects AI assistants to Microsoft Planner. Ask your AI assistant to create tasks, organise plans, manage buckets, and more, all through natural language.

> This is an unofficial, community/open-source MCP server for Microsoft Planner.

> It is not affiliated with, endorsed by, or sponsored by Microsoft.

Built with FastMCP and authenticated via Microsoft Entra ID (Azure AD) using the On-Behalf-Of (OBO) flow to call Microsoft Graph.

Table of Contents

• What Can It Do?
• Prerequisites
• Azure Entra ID Setup
• Installation
• Configuration
• Running the Server
• Connecting an MCP Client
• Available Tools
• Development
• Contributing
• License
• Security

What Can It Do?

Once connected, you can ask your AI assistant things like:

• *"Show me all my Planner tasks that are overdue"*
• *"Create a task called 'Prepare Q3 report' in the Marketing plan"*
• *"Move all incomplete tasks in the Sprint bucket to the Backlog bucket"*
• *"Mark the 'Update docs' task as complete"*

The AI assistant translates your request into the appropriate tool calls automatically.

*Note: This MCP only supports Planner basic tasks and plans*

Prerequisites

• A Microsoft 365 account with access to Microsoft Planner
• An Azure Entra ID (Azure AD) app registration (see Azure Setup below)
• An MCP-compatible client — for example:
  • VS Code
  • Claude Desktop

To run the server locally you also need:

• Python 3.12+
• uv (Python package manager)

Or, to run via Docker:

• Docker

Azure Entra ID Setup

An Azure app registration is required so the server can authenticate users and call Microsoft Graph on their behalf. You need admin access to an Azure Entra ID tenant (or ask your IT administrator).

Step 1 — Register the Application

1. Go to the Azure Portal → Microsoft Entra ID → App registrations → New registration

2. Enter a name (e.g. `Microsoft Planner MCP`)

3. Under Supported account types, choose the option appropriate for your organisation

4. Set the Redirect URI to Web → `http://localhost:8000/auth/callback`

5. Click Register

Step 2 — Configure API Permissions

1. In your new app registration, go to API permissions → Add a permission → Microsoft Graph → Delegated permissions

2. Add these permissions:

3. Click Grant admin consent for your organisation

Step 3 — Expose an API Scope

1. Go to Expose an API

2. Set the Application ID URI (accept the default `api://` or customise it)

3. Click Add a scope:

Step 4 — Set Token Version

1. Go to Manifest (or Authentication → Advanced settings in newer portal versions)

2. Set `"requestedAccessTokenVersion"` to `2`

3. Save

Step 5 — Create a Client Secret

1. Go to Certificates & secrets → New client secret

2. Add a description and choose an expiry period

3. Copy the Value immediately (it is only shown once)

Step 6 — Note Your IDs

You will need these three values for configuration:

Installation

Option A — Local (Python + uv)

git clone https://github.com/aixolotl/microsoft-planner-mcp
cd microsoft-planner-mcp
uv sync

Option B — Docker

No Python installation needed. See Running with Docker below.

Configuration

Copy the example environment file and fill in your Azure credentials:

cp .env.example .env

Edit `.env`:

# Azure App Registration (required)
CLIENT_ID=your-app-client-id
CLIENT_SECRET=your-app-client-secret
TENANT_ID=your-azure-tenant-id

# Public URL of this server (used for OAuth redirect URI)
BASE_URL=http://localhost:8000

# JSON-encoded list of allowed CORS origins
# Include http://localhost:6274 if using MCP Inspector for testing
ALLOWED_ORIGINS=["http://localhost:8000","http://localhost:6274"]

# Require FastMCP's extra client consent prompt per MCP client. (default: true)
# Prevents confused-deputy attacks by requiring users to explicitly approve each new
# client. Keep true in production; set false only for local dev with throwaway
# clients.
# REQUIRE_AUTHORIZATION_CONSENT=true

# Optional rate-limit overrides (per client sliding window)
# Defaults are 120 requests per 1 minute
# RATE_LIMIT_MAX_REQUESTS=120
# RATE_LIMIT_WINDOW_MINUTES=1

`REQUIRE_AUTHORIZATION_CONSENT` controls whether FastMCP prompts users to explicitly approve each new MCP client. Keep this `true` (default) in production to prevent confused-deputy attacks; set it to `false` only during local development with throwaway clients.

`RATE_LIMIT_MAX_REQUESTS` and `RATE_LIMIT_WINDOW_MINUTES` are optional and should stay commented out unless you need to override the defaults for your environment.

Running the Server

Local

uv run uvicorn src.server:app --host 0.0.0.0 --port 8000

The MCP endpoint is available at `http://localhost:8000/mcp`. A health check endpoint is at `http://localhost:8000/health`.

Running with Docker

# Pull the latest image
docker pull ghcr.io/aixolotl/microsoft-planner-mcp:latest

# Run with environment variables
docker run --rm -i \
  -e BASE_URL=https://localhost:8000 \
  -e CLIENT_ID=your_client_id \
  -e CLIENT_SECRET=your_api_token \
  -e TENANT_ID=your_tenant_id \
  -e ALLOWED_ORIGINS=["http://localhost:8000","http://localhost:6274", "http://localhost:3000"] \
  -e REQUIRE_AUTHORIZATION_CONSENT=true \
  ghcr.io/aixolotl/microsoft-planner-mcp:latest

This starts the MCP server on port 8000.

To override rate limits, also pass `-e RATE_LIMIT_MAX_REQUESTS=` and `-e RATE_LIMIT_WINDOW_MINUTES=` (defaults: `120` and `1`).

Running with Docker compose

docker compose up

This starts the MCP server on port 8000. The Docker Compose configuration also includes a Jaeger instance for trace visualisation (see OpenTelemetry Tracing).

Connecting an MCP Client

Once the server is running, configure your MCP client to connect to it.

VS Code

Add the following to your VS Code settings (`.vscode/settings.json` in your project, or your user settings):

{
  "mcp": {
    "servers": {
      "planner": {
        "type": "http",
        "url": "http://localhost:8000/mcp"
      }
    }
  }
}

Then use Copilot Chat in Agent mode and ask it to interact with your Planner tasks. Copilot will discover the available tools automatically.

Other MCP Clients

Any client that supports the Streamable HTTP transport can connect by pointing to `http://localhost:8000/mcp`. The server advertises OAuth metadata automatically — the client handles the authentication flow.

Available Tools

All tools are available to your AI assistant automatically once connected. You don't need to call them directly — just describe what you want in natural language. The parameter details below are provided for reference and for client developers.

Read-only tools are annotated with `readOnlyHint: true` so clients can skip confirmation prompts. Destructive tools (deletes) are annotated with `destructiveHint: true`.

User

`get_me`

Return the authenticated user's profile from Microsoft Graph.

• Parameters: None
• Returns: User profile object (id, displayName, mail, etc.) or `null`

`list_users`

Retrieve Microsoft 365 users by GUID, e-mail address, or free-text search. Useful for resolving the user GUIDs returned in task assignment objects to display names.

• Returns: List of user objects or `null`

> Note: When both `guids`/`emails` and `search` are supplied, the GUID/email filter takes priority. Very large lists of GUIDs or e-mail addresses are silently truncated to stay within the 2 048-character Graph URL limit.

Groups

`list_my_groups`

List all Microsoft 365 groups the authenticated user is a member of.

• Returns: List of group objects or `null`

> Note: The Groups tool will not return details like name or mail of groups with the standard permissions `Tasks.ReadWrite` documented here, *however* searching and filtering still works, so you can find a group with a specific name using this tool.

Plans

`list_my_plans`

List Planner plans shared with the authenticated user.

• Returns: List of plan objects or `null`

`list_group_plans`

List all Planner plans belonging to a Microsoft 365 group.

• Returns: List of plan objects or `null`

`create_plan`

Create a new Planner plan for a Microsoft 365 group.

• Returns: The created plan object or `null`

`delete_plan`

Delete a Planner plan.

• Returns: Confirmation message

`list_plan_categories`

Get category label definitions for a Planner plan. Returns all 25 category slots with their key (e.g. `category1`) and display name.

• Returns: List of category objects (`key`, `display_name`) or `null`

Buckets

`list_buckets`

List all buckets in a Planner plan.

• Returns: List of bucket objects or `null`

`create_bucket`

Create a new bucket in a Planner plan.

• Returns: The created bucket object or `null`

`delete_bucket`

Delete a Planner bucket.

• Returns: Confirmation message

Tasks

`list_my_tasks`

List all Planner tasks assigned to the authenticated user across all plans.

• Returns: List of task objects or `null`

`list_tasks`

List all tasks in a Planner plan.

• Returns: List of task objects or `null`

`get_task_details`

Get the full details for a task: description, checklist items, and external references.

• Returns: Task details object (description, checklist, references) or `null`

`create_task`

Create a new task in a Planner plan.

• Returns: The created task object or `null`

`update_task`

Update a task's standard fields and/or detail fields (description, checklist, references). Only provided fields are changed. When detail fields are specified, a separate API call updates the task details resource automatically.

• Returns: The updated task object, the updated task details object, or both (`{ "task": ..., "details": ... }`) depending on which fields were provided. Returns `null` if the result is empty.

> Note: The `update_task_details` tool has been removed. Use `update_task` with detail field parameters (`description`, `checklistItems`, `references`, `previewType`) instead. The task and details resources have separate ETags — provide `etagDetails` when available to avoid an extra round-trip, or omit it to let the tool auto-refresh.

`list_task_fields`

Return metadata for every field on a Planner task, including its data type, description, whether it is writable, and whether it requires a separate `get_task_details` call.

• Parameters: None
• Returns: List of field metadata objects (`name`, `type`, `description`, `writable`, `detailed`)

> Note: The list of fields is static as Planner Standard doesn't support custom fields. But this endpoint awaits Microsoft's future updates to the API to fully support Planner Premium.

`delete_task`

Delete a Planner task.

• Returns: Confirmation message

Development

This section covers building, testing, and contributing to the project.

Project Structure

src/
├── server.py                  # FastMCP app, middleware, route mounting
├── config.py                  # Settings via pydantic-settings
├── auth_provider.py           # Azure OAuth provider (OBO flow)
├── deps.py                    # Shared dependency helpers
├── graph_client_manager.py    # Singleton GraphClientManager with per-user OBO clients
├── telemetry.py               # OpenTelemetry setup
├── types.py                   # Shared structural types
├── services/
│   └── planner_service.py     # Business logic wrapping Graph SDK calls
└── tools/
    ├── me.py                  # get_me
    ├── groups.py              # list_my_groups
    ├── plans.py               # plan tools + list_plan_categories
    ├── tasks.py               # task tools + list_task_fields
    └── buckets.py             # bucket tools
tests/
├── conftest.py
├── test_buckets_tool.py
├── test_groups_tool.py
├── test_planner_service.py
├── test_plans_tool.py
└── test_tasks_tool.py

Installing Dev Dependencies

uv sync --dev

Running Tests

uv run pytest

With verbose output:

uv run pytest -v

Architecture

The server is built with FastMCP and uses these key patterns:

Server composition — Tools are split into five domain routers (`me`, `groups`, `plans`, `tasks`, `buckets`), each a standalone `FastMCP` instance mounted on the main app. This keeps each domain's tools, imports, and tests isolated.

Authentication — The server uses FastMCP's `OAuthProxy` pattern via a custom `AzureProvider`. Azure Entra ID does not support Dynamic Client Registration (DCR), so the provider acts as a DCR-compliant proxy facing MCP clients while using the pre-registered app credentials with Azure. When a tool call arrives, the server exchanges the MCP session token for a Microsoft Graph token via the On-Behalf-Of flow, scoped to `Tasks.ReadWrite` and `User.Read`.

Middleware — Five built-in middleware layers are stacked on the server (outermost first):

Client logging — Every tool sends real-time progress messages to the MCP client via `get_optional_context()`, so users see status updates like *"Fetching tasks…"* and *"Found 12 task(s)"* in their client.

OpenTelemetry Tracing

The server includes native OpenTelemetry instrumentation with zero overhead when unused. To enable trace export:

# Install the optional OTEL dependency group
uv sync --group otel

# Set the OTLP endpoint and start the server
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
export OTEL_SERVICE_NAME=microsoft-planner-mcp
uv run uvicorn src.server:app --host 0.0.0.0 --port 8000

Traces follow the MCP semantic conventions and work with any OTLP-compatible backend (Jaeger, Grafana Tempo, Datadog, New Relic, etc.). The Docker Compose setup includes Jaeger with a UI at `http://localhost:16686`.

Testing with MCP Inspector

MCP Inspector lets you test tools interactively in a browser:

1. Start the server:

uv run uvicorn src.server:app --host 0.0.0.0 --port 8000

2. In a separate terminal:

npx @modelcontextprotocol/inspector

3. Open the URL printed in the terminal (e.g. `http://localhost:6274/?MCP_PROXY_AUTH_TOKEN=...`)

4. Set Transport Type to `Streamable HTTP` and URL to `http://localhost:8000/mcp`, then click Connect.

Contributing

Contributions are welcome! Please open an issue or submit a pull request. See the issue templates for bug reports, feature requests, and tasks.

License

This project is licensed under the MIT License.