onenote-mcp
A secure MCP server for Microsoft OneNote via the Graph API, enabling listing/reading notebooks, sections, and pages, with optional gated write tools. Read-only by default and deployable to Azure Container Apps.
README
OneNote MCP server
A secure remote MCP server around the Microsoft Graph OneNote API (v1.0).
Serves an MCP endpoint over Streamable HTTP that claude.ai (as a custom connector),
claude mcp, ChatGPT, or the MCP Inspector can register. Modeled on the sibling
lawmatics-mcp and deploys the same way (Azure Container App).
- Read-only by default. Write tools (
create_page,append_to_page) are registered only whenENABLE_WRITES=true, and each additionally requires an explicitconfirm: true(a dry-run preview is returned otherwise). - No secrets in source. All credentials come from env / Azure Key Vault.
- No content in logs. Titles, page HTML, and tokens are never logged. See
src/logging.ts.
Tools
| Tool (registered name) | Requested name | Purpose |
|---|---|---|
onenote_list_notebooks |
list_notebooks |
List notebooks for the configured user/group/site |
onenote_list_sections |
list_sections(notebookId) |
List sections in a notebook |
onenote_list_pages |
list_pages(sectionId) |
List pages in a section (newest first) |
onenote_read_page |
read_page(pageId) |
Return a page's HTML/text content |
onenote_create_page |
create_page(sectionId, title, htmlBody) |
Create a page (write; gated) |
onenote_append_to_page |
append_to_page(pageId, htmlBody) |
Append an HTML block to a page (write; gated) |
Tools are prefixed with onenote_ for discoverability (house convention), matching how
lawmatics-mcp prefixes its tools. Page content is HTML throughout.
Surgical edits.
append_to_pageis a coarse whole-body append. To change a specific existing element, you mustPATCH /onenote/pages/{id}/contentwith a command whosetargetis that element'sdata-id. Those anchors only exist if you first read the page withincludeIds=true(onenote_read_pagedoes this by default). This is documented insrc/onenoteClient.tsas an extension point; no destructive edit/replace tool is shipped by design.
(a) Azure app registration + admin consent
You need one app registration. Choose application (app-only, client credentials) or delegated (on behalf of a signed-in user). Application mode is the primary/default flow.
1. Register the app
- Entra admin center → Microsoft Entra ID → App registrations → New registration.
- Name:
onenote-mcp. Supported account types: Single tenant (typical). Register. - From Overview, copy the Directory (tenant) ID →
GRAPH_TENANT_IDand the Application (client) ID →GRAPH_CLIENT_ID.
2. Add a client secret
- Certificates & secrets → Client secrets → New client secret → set an expiry.
- Copy the secret Value (not the Secret ID) →
GRAPH_CLIENT_SECRET. Store it in Key Vault.
3. Add the Graph permission + grant admin consent
Application mode (client credentials — default):
- API permissions → Add a permission → Microsoft Graph → Application permissions.
- Add
Notes.ReadWrite.All. - Click Grant admin consent for <tenant> (a Global Admin must do this). Status must show a green check.
- Set
GRAPH_AUTH_MODE=application,ONENOTE_TARGET_TYPE=user, andONENOTE_TARGET_ID=<UPN or object id>(app-only has no "me" — you must name whose notebooks to act on;group/sitealso work).
Delegated mode (on behalf of a user):
- API permissions → Add a permission → Microsoft Graph → Delegated permissions.
- Add
Notes.ReadWriteandoffline_access(andUser.Read). Grant admin consent (or let the user consent at first sign-in). - Authentication → Add a platform → Web → redirect URI
http://localhost:8082/oauth/callback(add your production callback too if you script re-auth there). Save. - Set
GRAPH_AUTH_MODE=delegatedand mint a refresh token (see below);ONENOTE_TARGET_TYPEdefaults tome.
4. (Delegated only) mint the refresh token — one time
Access tokens are short-lived; offline_access yields a refresh token the server caches and rotates.
# set GRAPH_TENANT_ID / GRAPH_CLIENT_ID / GRAPH_CLIENT_SECRET first (see .env.example)
npm run authorize
Sign in and approve; copy the printed refresh token into your secret manager as
GRAPH_REFRESH_TOKEN. The script uses PKCE and never writes the token to disk.
(b) Run locally
Windows/OneDrive note: don't
npm installinside a OneDrive-synced folder. Copy the project to a local path first (e.g.C:\dev\onenote-mcp), install/build there, and keep only source in OneDrive.
cp .env.example .env # fill via your secret manager
npm install
npm run build
npm start # serves POST http://localhost:8082/mcp (health: GET /healthz)
Quick smoke test with the MCP Inspector:
npx @modelcontextprotocol/inspector
# Transport: Streamable HTTP · URL: http://localhost:8082/mcp
# Header: Authorization: Bearer <MCP_AUTH_TOKEN>
Required env: MCP_AUTH_TOKEN, GRAPH_TENANT_ID, GRAPH_CLIENT_ID, GRAPH_CLIENT_SECRET,
plus ONENOTE_TARGET_ID (application mode) or GRAPH_REFRESH_TOKEN (delegated mode).
Writes are off unless ENABLE_WRITES=true.
(c) Deploy as an Azure Container App
Same pattern as lawmatics-mcp. Secrets live in Container Apps secrets (optionally sourced from
Azure Key Vault); nothing is baked into the image.
# variables
RG=ctg-mcp-rg
LOC=eastus
ACR=ctgmcpacr # must be globally unique
ENV=ctg-mcp-env
APP=onenote-mcp
# 0) resource group
az group create -n $RG -l $LOC
# 1) build & push the image with ACR (no local Docker needed)
az acr create -n $ACR -g $RG --sku Basic --admin-enabled true
az acr build -r $ACR -t onenote-mcp:latest .
# 2) Container Apps environment
az containerapp env create -n $ENV -g $RG -l $LOC
# 3) create the app (ingress 8082) with secrets
az containerapp create \
-n $APP -g $RG --environment $ENV \
--image $ACR.azurecr.io/onenote-mcp:latest \
--registry-server $ACR.azurecr.io \
--target-port 8082 --ingress external \
--min-replicas 1 --max-replicas 3 \
--secrets \
mcp-auth-token=$MCP_AUTH_TOKEN \
graph-client-secret=$GRAPH_CLIENT_SECRET \
--env-vars \
PORT=8082 \
MCP_PATH=/mcp \
MCP_AUTH_TOKEN=secretref:mcp-auth-token \
GRAPH_AUTH_MODE=application \
GRAPH_TENANT_ID=$GRAPH_TENANT_ID \
GRAPH_CLIENT_ID=$GRAPH_CLIENT_ID \
GRAPH_CLIENT_SECRET=secretref:graph-client-secret \
ONENOTE_TARGET_TYPE=user \
ONENOTE_TARGET_ID=$ONENOTE_TARGET_ID \
ENABLE_WRITES=false
# 4) get the public URL — your MCP endpoint is https://<fqdn>/mcp
az containerapp show -n $APP -g $RG --query properties.configuration.ingress.fqdn -o tsv
To enable writes later: az containerapp update -n $APP -g $RG --set-env-vars ENABLE_WRITES=true.
Key Vault (recommended): store MCP_AUTH_TOKEN, GRAPH_CLIENT_SECRET, and (delegated)
GRAPH_REFRESH_TOKEN in Key Vault, give the app a managed identity with Key Vault Secrets User,
and reference them via --secrets mcp-auth-token=keyvaultref:<secret-uri>,identityref:<identity-id>.
(d) Connect it
As a claude.ai connector
- Deploy (above) and confirm
GET https://<fqdn>/healthzreturns{"status":"ok"}. - In claude.ai → Settings → Connectors → Add custom connector:
- URL:
https://<fqdn>/mcp - Authentication: the connector must send
Authorization: Bearer <MCP_AUTH_TOKEN>.
- URL:
- Save; the six OneNote tools appear (writes only if
ENABLE_WRITES=true).
Via claude mcp (Claude Code)
claude mcp add onenote --transport http https://<fqdn>/mcp \
--header "Authorization: Bearer <MCP_AUTH_TOKEN>"
Or local dev against http://localhost:8082/mcp. Verify with claude mcp list, then /mcp in-session.
Graph facts encoded here
- Base
https://graph.microsoft.com/v1.0; bearer auth; OData$select/$orderby/$top; list envelope{ value: [...], "@odata.nextLink" }. - OneNote lives under a user/group/site context:
/me/onenote(delegated) or/users/{id}/onenote(application). App-only cannot use/me. - Page content is HTML. Create =
POST .../sections/{id}/pages(Content-Type: text/html, title from<title>); append =PATCH .../pages/{id}/contentwith a JSON command array; read =GET .../pages/{id}/content?includeIDs=true(returns HTML withdata-idanchors). - Client-credentials access tokens (~60–90 min) are cached in memory and re-acquired ~1 min before
expiry; delegated refresh tokens are used and rotated in memory. See
src/graphAuth.ts.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.