microsoft-tasks-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@microsoft-tasks-mcp what do I have to do today across To Do and Planner?"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-server-microsoft-tasks
In one sentence: an MCP server that lets AI coding agents read your Microsoft Planner + Microsoft To Do tasks across all M365 tenants you sign into, without bypassing Microsoft's auth and without ever modifying tasks the agent didn't create itself.
What is this for?
You work across multiple Microsoft 365 tenants — consultancy, customer engagements, your own org. Tasks are scattered across:
Microsoft To Do — your personal task lists, the place flagged emails land, the place ad-hoc reminders go.
Microsoft Planner — group-scoped boards, one per M365 group / Team, where collaborative work lives.
Surfacing "what do I have to do today" already requires the user to mentally union both surfaces. Worse, popular AI agents that can talk to Microsoft 365 either:
bypass Microsoft's modern auth (broken attribution),
can't see the multi-tenant boundary,
or auto-modify tasks created by other people (terrifying).
mcp-server-microsoft-tasks fixes all three: local process per tenant, multi-profile, Microsoft Graph for full attribution, read-only by default, writes opt-in (v0.2+), agent-created-only. The per-profile registry is the hard gate — write tools refuse to touch any task whose ID isn't in the registry of "tasks this profile created".
Sister project to mcp-server-sharepoint and mcp-server-outlook. Same authorship pattern, same auth shape (mcp-microsoft-graph-auth), different surface.
Related MCP server: todo-mcp
Installation
pip install mcp-server-microsoft-tasks
# or, with uv (recommended):
uv tool install mcp-server-microsoft-tasks
# or, on the fly without installing globally:
uvx mcp-server-microsoft-tasks --helpRequires Python 3.11+. Works on Linux, macOS, Windows.
Quickstart
1. Sign in once (out of band)
uvx mcp-server-microsoft-tasks login --account-type work_or_school
# or, for a personal Microsoft account:
uvx mcp-server-microsoft-tasks login --account-type personalThe --account-type flag is required: work_or_school for any Microsoft 365 tenant account (both Planner and To Do work), personal for outlook.com / hotmail.com / live.com / msn.com (only To Do works — Planner needs a work/school M365 group). Internally this picks the right Microsoft Identity Device Code authority (/organizations vs /consumers); TASKS_TENANT_ID is kept as a power-user / CI escape hatch.
The output renders the device code first in its own code block, the URL second on its own line. Copy the code, click the link, paste, sign in. Your refresh token is cached locally (OS keyring on macOS / Windows / Linux desktop; encrypted-file fallback otherwise). The MCP server itself never blocks for human interaction afterwards.
When used through an MCP tool (tasks_login_begin), the agent receives the same account_type requirement via tool-schema description; calling without it returns a structured LoginAccountTypeRequiredError that instructs the agent to ask the user which account kind to sign in with.
2. Wire it into Claude Code (or any MCP client)
In your project's .mcp.json:
{
"mcpServers": {
"microsoft-tasks": {
"command": "uvx",
"args": ["mcp-server-microsoft-tasks"]
}
}
}Restart Claude Code. The agent now has the full v0.1 read surface (see below). All v0.1 tools are read-only — the consent screen on sign-in shows Tasks.Read and Group.Read.All, no write scopes.
3. Multi-tenant: sign in to a second profile
uvx mcp-server-microsoft-tasks login --profile customer-acmeThen run a second MCP server process pointed at that profile:
{
"mcpServers": {
"microsoft-tasks-acme": {
"command": "uvx",
"args": ["mcp-server-microsoft-tasks"],
"env": { "TASKS_PROFILE": "customer-acme" }
}
}
}Each profile is one tenant. No cross-tenant token sharing.
Use case — a daily-shape pulse
Concrete dialogue showing the headline workflow on the v0.1 read surface:
You: What do I have to do today across To Do and Planner?
Agent: [calls tasks_assigned_to_me(include_completed=False, limit=20)]
Five items on your plate, sorted by due date:
Today
- "Review draft contract for Acme" (Planner / Sprint 5 / Todo bucket)
due today, 17:00 — assigned to you + 1 other
- "Pick up dry cleaning" (To Do / Personal)
no time set
Tomorrow
- "Send follow-up to vendor" (Planner / Customer ops / Doing bucket)
due tomorrow, 09:00
Later this week
- "Renew passport" (To Do / Personal) — Friday
- "ISO 27001 control review" (Planner / Compliance / Todo)
Sunday — high priority
You: What's the deal with the ISO control review? Search my tasks for "27001".
Agent: [calls tasks_search("27001", source="all")]
Three matches — one Planner, two To Do:
- "ISO 27001 control review" (Planner / Compliance, due Sunday)
→ bucket "Todo", priority high, no checklist yet
- "Read ISO 27001:2022 changes" (To Do / Reading list, no due date)
- "Schedule auditor call re: ISO 27001 surveillance" (To Do, completed last week)
Want me to pull the full body and checklist of the Planner one?
You: Yes please.
Agent: [calls planner_task_get(task_id="...", include_details=True)]
"ISO 27001 control review" — full description:
Quarterly review of all 93 controls in Annex A. Walk through
the SoA spreadsheet, flag any controls whose ownership has
changed, attach evidence links from SharePoint.
Checklist (4 of 12 done):
- [x] Pull current SoA from SharePoint
- [x] Identify controls touching authentication
- [ ] Map to MFA rollout milestone (still open)
- [ ] ...Every action above is a read — no Planner/To-Do state was modified. The server's default install requests no write scopes; the consent screen reads "this app can read your tasks", not "modify". Write tools (create / update / complete / delete) land in v0.2 behind the explicit TASKS_ALLOW_WRITES=true env flag.
Use case — drafting follow-up tasks after a meeting
A second dialogue showing v0.2 write tools (TASKS_ALLOW_WRITES=true). Note that the agent only ever creates tasks; it never marks anyone else's task done.
You: I just got out of the steering meeting with the ISMS team. Three follow-ups
— please add them to the Compliance Planner plan, Todo bucket. Owner:
me. Due Friday. The items are:
1. Update the Statement of Applicability for control A.5.1
2. Schedule the Q3 surveillance audit pre-call with the auditor
3. Send the policy-review reminder to the legal team
Agent: [calls planner_plans()]
[calls planner_buckets("Compliance-plan-id") to find the Todo bucket]
[calls planner_task_create x3, each with plan_id + bucket_id + title +
due_date + assignees=["d.koller@xmv.de"]]
Created three Planner tasks in Compliance / Todo, all due Friday:
- "Update SoA for A.5.1" — task id task-abc1
→ https://tasks.office.com/<tenant>/Home/Task/task-abc1
- "Schedule Q3 surveillance pre-call with auditor" — task-abc2
→ https://tasks.office.com/<tenant>/Home/Task/task-abc2
- "Send policy-review reminder to legal" — task-abc3
→ https://tasks.office.com/<tenant>/Home/Task/task-abc3
All three are now in this MCP profile's registry, so I can update or
complete them later if you ask. Want me to also drop a "review the SoA
edits" reminder on your personal To Do list?
You: Yes please, due Thursday.
Agent: [calls todo_task_create(list_id=<default>, title=..., due_date="2026-08-13",
importance="high")]
Added to your default Tasks list:
- "Review the SoA edits before sending" — due Thursday, importance high
You: Mark task-abc3 as complete — I already pinged legal directly.
Agent: [calls planner_task_complete("task-abc3")]
Marked "Send policy-review reminder to legal" complete. The task was in
my registry (I created it earlier this conversation), so the action went
through. If you'd asked me to mark a task you or a colleague had created,
I would have refused — only my own tasks are eligible.The last exchange shows the load-bearing safety guarantee in action. The per-profile registry on disk (~/.cache/mcp-server-microsoft-tasks/<profile>/tasks.json) lists every task this MCP server created. planner_task_update, _complete, and _delete consult that registry before hitting Microsoft Graph and refuse with NOT_OWNED_BY_PROFILE if the task isn't there. ETag-based optimistic concurrency adds a second guard against silently clobbering external edits — see docs/app-concept.md § Conflict / safety semantics.
v0.1 tool surface (read-only)
Tool | What it does |
| Non-blocking Device Code login as MCP tools — agent surfaces code + URL without leaving the chat. |
| Enumerate / fetch Microsoft To Do lists (default Tasks list, flagged-emails list, user-created lists). |
| List / fetch tasks within a To Do list. |
| Enumerate Planner plans across the user's M365 groups (or within one group via |
| List buckets (columns) within a Planner plan. |
| List / fetch Planner tasks. |
| Unified across To Do + Planner. Sorted by due date ascending. |
| Cross-source substring search; |
| Incremental Planner diff since last poll — returns |
Every tool returns a unified task envelope with id, title, status, due_date, assignees, web_url, source, etag, plus source-specific extras (list_id / body_preview / categories / importance / reminder_date for To Do; plan_id / bucket_id / priority / percent_complete / applied_categories for Planner). Agents can route follow-up calls correctly off the source tag without learning two response shapes.
Incremental polling with tasks_changes_since
tasks_changes_since(scope, max_results=200) lets an agent detect what changed since it last looked — without re-processing every task on every poll.
Pass a scope dict to control which tasks are polled:
Scope | What gets polled |
| All tasks in one Planner plan ( |
| Tasks assigned to the signed-in user ( |
| Only tasks this MCP profile created — one |
The tool returns:
{
"added": [<task envelope>, ...],
"modified": [<task envelope>, ...],
"removed": [{"id": "...", "last_known_title": null}, ...],
"cursor_advanced": true
}On the first call for a scope, every visible task is returned as added and the cursor is initialised. On subsequent calls, only tasks that appeared, had their lastModifiedDateTime advance, or disappeared from the response are returned. Everything else comes back empty.
The cursor file lives at ~/.cache/mcp-server-microsoft-tasks/<profile>/cursors.json (mode 0o600) and is updated atomically on every call. Each scope is tracked independently (keyed by sha256 of the JSON-serialised scope). The last_modified_max watermark is monotonic — a temporarily stale Graph timestamp can never roll the cursor backwards.
v0.2 — write tools, opt-in via TASKS_ALLOW_WRITES=true
Tool | What it does |
| Writes on To Do tasks — only tasks this profile's registry created. |
| Writes on Planner tasks — same registry guarantee. |
| Attach / detach an HTTP/HTTPS URL reference (OneNote, SharePoint, etc.) on a profile-owned Planner task. Same registry guarantee. |
| Inspect this profile's "I created this" registry. |
To enable, set TASKS_ALLOW_WRITES=true in the MCP client config (e.g. via env block in .mcp.json):
{
"mcpServers": {
"microsoft-tasks": {
"command": "uvx",
"args": ["mcp-server-microsoft-tasks"],
"env": { "TASKS_ALLOW_WRITES": "true" }
}
}
}The default install does NOT request Tasks.ReadWrite. Setting TASKS_ALLOW_WRITES=true adds the scope at sign-in time AND registers the write tools at MCP-server-start time.
The two write-time safety guarantees
Per-profile registry on disk records every task this server created (
~/.cache/mcp-server-microsoft-tasks/<profile>/tasks.json, mode 0o600). Write tools refuse — at the tool layer, before any Microsoft Graph call — to act on tasks not in the registry. The error isNOT_OWNED_BY_PROFILE. Hand-created tasks in Microsoft Planner / To Do are never modified by the agent; tasks created by other agents (different MCP profile, different process, different machine) are likewise untouchable.ETag-based optimistic concurrency via
If-Match. The registry stores the last ETag this server saw; every PATCH / DELETE attaches it; Microsoft Graph returns 412 Precondition Failed if the task changed externally between the agent's read and the write. The MCP surfaces this asEXTERNALLY_MODIFIEDso the agent re-fetches and decides.
No bulk operations, no auto-assignment to other users, no plan/list creation: each write tool acts on exactly one task per call, and assignees on planner_task_create is filled only from values the human typed in chat. See docs/app-concept.md § Conflict / safety semantics.
Non-admin tenants — opt out of Planner with MS_TASKS_NO_PLANNER=true
Microsoft Planner requires the Group.Read.All Graph scope to enumerate the M365 groups that own plans. That scope is admin-consent in most tenants — a non-admin user signing in for the first time will hit a "your administrator must approve this app" prompt.
If you only care about your personal Microsoft To Do tasks, opt out of Planner entirely:
{
"mcpServers": {
"microsoft-tasks": {
"command": "uvx",
"args": ["mcp-server-microsoft-tasks"],
"env": { "MS_TASKS_NO_PLANNER": "true" }
}
}
}With this flag set:
The OAuth scope request drops
Group.Read.All, so the consent screen no longer needs admin approval.The MCP server skips registering all
planner_*tools at start-up (so the agent never sees them).tasks_assigned_to_meandtasks_searchsilently exclude the Planner half — they still work, just on the To Do side only.
Both the writes opt-in (TASKS_ALLOW_WRITES=true) and the no-Planner opt-out are independent: you can have writes-with-Planner, writes-without-Planner, reads-with-Planner, or reads-without-Planner.
Recurring Planner tasks — opt in to /beta with MS_TASKS_PLANNER_BETA=true
Microsoft Graph's Planner recurrence APIs are /beta-only. To create a recurring task — or to see the recurrence field on read — opt in to the beta endpoint:
{
"mcpServers": {
"microsoft-tasks": {
"command": "uvx",
"args": ["mcp-server-microsoft-tasks"],
"env": { "TASKS_ALLOW_WRITES": "true", "MS_TASKS_PLANNER_BETA": "true" }
}
}
}With the flag on, every Planner tool routes through /beta/planner/.... planner_task_create and planner_task_update accept an optional recurrence argument; the unified envelope surfaces recurrence on read with the schedule + series-tracking metadata Graph populates.
Example agent call (creating a weekly status-update task):
{
"tool": "planner_task_create",
"arguments": {
"plan_id": "...", "bucket_id": "...", "title": "Weekly status",
"recurrence": {
"schedule": {
"patternStartDateTime": "2026-05-11T08:00:00Z",
"pattern": {
"type": "weekly", "interval": 1,
"daysOfWeek": ["monday"], "firstDayOfWeek": "sunday"
}
}
}
}
}To stop a series: planner_task_update(task_id, recurrence={"schedule": null}). (Graph rejects setting top-level recurrence to null on a task that already has it.) Without the flag, passing a recurrence argument raises before the HTTP call with a clear pointer to MS_TASKS_PLANNER_BETA.
Token storage
mcp-server-microsoft-tasks uses mcp-microsoft-graph-auth (sister library) to manage tokens. Default backend on macOS / Windows / Linux-desktop is the OS keyring; on headless Linux the fallback is a 0600 plain file at ~/.cache/mcp-server-microsoft-tasks/<profile>/token.json. For CI / encrypted-file mode, set MS_TASKS_TOKEN_PASSPHRASE and MS_TASKS_TOKEN_STORE=encrypted-file. Override the auto-pick with MS_TASKS_TOKEN_STORE=keyring|file|encrypted-file.
Documentation
Document | Description |
Vision, tool surface, auth model, conflict semantics, Testability section | |
XMV's project-agnostic baseline (test layers, source-control, PR discipline) | |
Brief for AI coding agents — project facts, tech stack, behaviour rules | |
What the OAuth app sees, what XMV sees (nothing), GDPR pointers | |
Default OAuth app, BYO override, disclaimer | |
Contribution flow | |
Vulnerability disclosure | |
Keep-a-changelog history |
Licence
Dual-licensed under either of:
Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
MIT License (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this project by you, as defined in the Apache-2.0 licence, shall be dual-licensed as above, without any additional terms or conditions.
Contact
Organisation: XMV Solutions GmbH
Email: oss@xmv.de
Website: https://xmv.de/en/oss/
GitHub: @XMV-Solutions-GmbH
Available Tools
14 toolsplanner_bucketsARead-onlyIdempotent
List buckets (columns) within a Planner plan. Each bucket has id, name, plan_id, order_hint, etag. The order_hint follows Microsoft Graph's lexicographic ordering scheme (read-only — buckets ship pre-ordered as the user arranged them in Planner). Read-only.
Inputs: plan_id (required). Returns: result
planner_plan_getARead-onlyIdempotent
Fetch one Microsoft Planner plan by id. Same shape as an entry returned by planner_plans. Read-only.
Inputs: plan_id (required). Returns: object
planner_plansARead-onlyIdempotent
List Microsoft Planner plans the signed-in user can see. Without group_id, enumerates the user's M365 groups via /me/memberOf (requires Group.Read.All admin-consent — already granted on the XMV-published OAuth app) and aggregates plans across them. With group_id, lists plans within that single group. Each plan has id, title, owner_group_id, created_date_time, etag. Read-only.
Inputs: group_id, limit. Returns: result
planner_task_getARead-onlyIdempotent
Fetch one Planner task by id. Returns the unified task envelope (same shape as items in planner_tasks). Pass include_details=True to additionally fetch the task's description, checklist, references, and preview_type (one extra Graph round-trip to /planner/tasks/{id}/details). Read-only.
Inputs: task_id (required), include_details. Returns: object
planner_tasksARead-onlyIdempotent
List tasks within a Planner plan. Returns each task in the unified envelope: id, title, status (completed/not_completed — derived from percentComplete >= 100), due_date, assignees (list of M365 user-ids assigned to the task), web_url (deep-link to tasks.office.com built from the access token's tid claim), source ('planner'), etag, plan_id, bucket_id, priority, percent_complete, applied_categories, created_date_time, last_modified_date_time. Optionally narrow by bucket_id and status_filter ('all'/'completed'/'not_completed').
Inputs: plan_id (required), bucket_id, status_filter, limit. Returns: result
tasks_assigned_to_meARead-onlyIdempotent
Cross-source view: every Microsoft To Do task in the user's lists plus every Microsoft Planner task assigned to the user, merged into one list. Sorted by due_date ascending (None last). include_completed=False excludes completed tasks from both surfaces. Each entry is a unified task envelope tagged with source ('todo' or 'planner') and profile so the agent can route follow-up calls correctly. Optional profiles=[...] fans out across multiple signed-in tenants and merges; per-profile failures are best-effort skipped and listed in _skipped_profiles. Returns {'tasks': [...], '_skipped_profiles': [...]}. Read-only.
Inputs: include_completed, limit, profiles. Returns: object
tasks_changes_sinceARead-only
Incremental diff of Microsoft Planner tasks since the last call. Polls Graph, compares against an on-disk cursor, and returns added, modified, removed envelopes plus cursor_advanced (bool). scope controls which tasks are polled — pass one of: {"kind": "plan", "plan_id": "..."} (all tasks in a plan), {"kind": "assigned_to_me"} (tasks assigned to the signed-in user), or {"kind": "registry"} (only tasks this MCP profile created — one GET per registry id). Each scope is tracked independently via a sha256 cursor key. First call → everything returned as added, cursor initialised. Subsequent calls return only tasks that appeared, changed (lastModifiedDateTime advanced), or disappeared since the previous poll. removed entries carry {"id": "...", "last_known_title": null} — the title is not available because the task is gone. Cursor file: ~/.cache/mcp-server-microsoft-tasks/<profile>/cursors.json, mode 0o600. Read-only on Graph; mutates only the local cursor file.
Inputs: scope (required), max_results. Returns: object
tasks_login_beginAIdempotent
Drive the OAuth Device Code flow as an MCP tool. Returns immediately, non-blocking. Surfaces user_code + verification_url so the agent can show them; polls Microsoft Identity in the background until the user completes sign-in OR the device code expires (~15 min cap). The agent then polls tasks_login_status until it flips to signed_in (or to a terminal expired / failed). REQUIRED parameter account_type: pass 'personal' for outlook.com / hotmail.com / live.com / msn.com (Microsoft To Do works; Planner does NOT — requires a work/school M365 group), or 'work_or_school' for any Microsoft 365 tenant account incl. B2B guests (both To Do and Planner work). The choice determines which Microsoft Device Code landing page the user is sent to (https://www.microsoft.com/link vs https://login.microsoft.com/device) — there is no auto-detection before sign-in. If you don't know, ASK THE USER first. Calling without account_type returns a structured error explicitly instructing you to elicit the choice. Idempotent: a non-expired pending session for the profile is returned as-is unless force=True. force=True cancels the in-flight session and starts a fresh flow. Returns the session's public view: session_id, user_code, verification_url, verification_url_complete, expires_at, time_remaining_s, status, signed_in_user_upn, error. AGENT_INSTRUCTIONS: Present the verification code to the user inside a fenced code block (so it can be copied with one click) and present the verification URL as a plain markdown link on its own line. Do not paraphrase, do not embed the code inside prose, do not wrap the URL in bold. Example:
Code:
```
ABCD-1234
```
Sign-in URL: https://www.microsoft.com/linkRationale: in a chat UI, a code inside a fenced block gets a one-click copy button; a bare URL becomes clickable; bold-wrapped links and inline codes do not.
Inputs: account_type, force. Returns: object
tasks_login_statusARead-onlyIdempotent
Return the current Microsoft 365 sign-in status for this profile. Three states: signed_in (a usable token exists, regardless of how it got there — CLI login, tasks_login_begin tool, even days ago), pending (a Device Code flow is in flight from a recent tasks_login_begin call; the response carries user_code + verification_url so the agent can re-display the prompt), or none (no token, no flow — the agent should call tasks_login_begin). Read-only: actively probes the token store + does at most one /me round-trip on a fresh signed_in to learn the UPN. AGENT_INSTRUCTIONS: When status='pending', present the verification code to the user inside a fenced code block (so it can be copied with one click) and present the verification URL as a plain markdown link on its own line. Do not paraphrase, do not embed the code inside prose, do not wrap the URL in bold. Same format as tasks_login_begin.
Inputs: no named parameters. Returns: object
tasks_searchARead-onlyIdempotent
Case-insensitive substring search across the user's To Do tasks and Planner tasks. Matches against title and body_preview. source narrows to a single surface — 'all' (default), 'todo', or 'planner'. Returns up to limit matches in the unified envelope shape. Read-only. Note: implementation is client-side because neither surface exposes a server-side $search for tasks; performance is fine at typical task volumes (hundreds, not hundreds of thousands).
Inputs: query (required), source, limit. Returns: result
todo_list_getARead-onlyIdempotent
Fetch a single Microsoft To Do list by id. Same shape as an entry returned by todo_lists. Read-only.
Inputs: list_id (required). Returns: object
todo_listsARead-onlyIdempotent
List the signed-in user's Microsoft To Do lists. Each list has id, display_name, is_owner, is_shared, well_known_list_name (e.g. 'defaultList' for the built-in Tasks list, 'flaggedEmails' for the Outlook flagged-mails list, or None for user-created lists), and etag. Read-only — does not modify anything. To list tasks within a list, pass the returned id to todo_tasks.
Inputs: limit. Returns: result
todo_task_getARead-onlyIdempotent
Fetch one Microsoft To Do task by id within its list. Both list_id and task_id are required — Microsoft Graph has no global task-by-id endpoint for To Do. Returns the unified task envelope (same shape as items in todo_tasks). Read-only.
Inputs: list_id (required), task_id (required). Returns: object
todo_tasksARead-onlyIdempotent
List tasks in a Microsoft To Do list. Returns each task in the unified envelope: id, title, status (completed/not_completed), due_date, assignees (empty for To Do — per-user surface), web_url (None for To Do — no public deep-link), source (always 'todo'), etag (for write concurrency), list_id, body_preview, categories, importance, reminder_date, is_reminder_on, last_modified_date_time, created_date_time. status_filter defaults to 'all'; pass 'completed' or 'not_completed' to narrow.
Inputs: list_id (required), status_filter, limit. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 14 tools
Each tool targets a distinct resource or action: Planner plans/buckets/tasks, To Do lists/tasks, cross-source views, search, incremental changes, and authentication. No overlap in purpose.
All tools follow a consistent verb_noun pattern with underscores (e.g., planner_plan_get, tasks_assigned_to_me, todo_lists), making them predictable for an agent.
14 tools is well-scoped for a task management server covering two surfaces (Planner and To Do) plus cross-source views and authentication. Each tool earns its place.
The tool set is almost entirely read-only; no create, update, or delete operations for tasks or lists are provided. This is a significant gap for a task management server, limiting agents to observation only.
Maintenance
Related MCP Connectors
Task management for people and AI agents, with scoped OAuth access to issues, projects, and docs.
Task management for people and AI agents, with scoped OAuth access to issues, projects, and docs.
Shared task layer for AI agents: tasks, handoffs and decisions that persist across tools.
AI-native task management: list, create, update and archive tasks with rich context for AI agents
Related MCP Servers
- AlicenseBqualityDmaintenanceEnables interaction with Microsoft Planner tasks through natural language using Azure CLI authentication. Supports listing plans, creating/updating/deleting tasks, managing buckets, and integrating GitHub links without complex OAuth setup.911ISC
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to manage Microsoft To Do tasks via the Graph API, supporting multiple accounts, encrypted auth, and full CRUD operations.2-
- AlicenseNot gradedqualityCmaintenanceEnables coding agents to manage Microsoft To Do lists and tasks through Microsoft Graph, with optional support for checklist items, linked resources, attachments, and delta synchronization.1MIT
- AlicenseAqualityBmaintenanceEnables AI agents to read and edit Microsoft To Do tasks, lists, and steps via Microsoft Graph, including due dates, reminders, and subtasks.121MIT