Introducing the Beyond Time MCP Server: Track Goals From Claude, Cursor, and Any AI Client

For a decade, "AI-powered productivity" meant opening yet another app and copy-pasting model output somewhere useful. The interface was the bottleneck. Goals in one place, your AI assistant in another, your calendar in a third, and the only thing connecting them was you.

The Beyond Time MCP server is live at https://beyondtime.ai/mcp. If you use Claude Desktop, Claude Code, Cursor, or any other client that speaks the Model Context Protocol, you can now create goals, plan milestones, schedule actions, build habits, and ask for daily reflections without leaving your AI client. Your goal system becomes a tool the AI can actually use, not a separate tab you have to remember to open.

This post covers what shipped and how to connect your client in five minutes.

What is the Beyond Time MCP server?

The Beyond Time MCP server is a remote, stateless implementation of the Model Context Protocol that exposes your Beyond Time account as a set of structured tools to any MCP-compatible AI client. The endpoint lives at https://beyondtime.ai/mcp and uses OAuth-protected bearer tokens to authenticate requests.

Concretely, an AI client can call tools like beyondtime_create_goal, beyondtime_list_milestones, or beyondtime_complete_action and operate on your real data: the same goals you see in the app, the same milestones, the same habits and routines you've already configured. There is no separate "AI account" or sandbox. The MCP server is a new face on the same Beyond Time you already use.

Each request is fully stateless: a per-request MCP server is spun up, your bearer token is verified, your Supabase session is resolved, the tool runs against your account, and the server is discarded. That keeps the security model boring, which is what you want from something that has write access to your goal system.

You don't need to install a binary or run a local proxy. If your AI client supports remote MCP servers, you point it at the URL, complete an OAuth flow once, and you're done.

What is Model Context Protocol and why does it matter for goal tracking?

Model Context Protocol, or MCP, is an open specification for connecting AI clients to external tools and data. Anthropic introduced it in late 2024, and it has since become the closest thing the AI ecosystem has to a standard plug for "let the model do useful things in the real world."

Before MCP, every AI integration was bespoke. If you wanted Claude to write to your task manager, someone built a Claude-specific integration. Cursor talking to the same task manager was a separate one. The result was fragile one-off connectors, mostly broken.

MCP collapses that into a single contract. A server exposes tools using a defined schema. Any client that speaks MCP can discover and call them. Boring on purpose.

For goal tracking, MCP unlocks something interesting. Goal management is structurally a great fit for AI assistance: nested objects (goals, milestones, actions), recurring rituals (planning, reflecting), and natural-language inputs ("I want to run a half marathon by October"). The bottleneck has never been the AI's reasoning. It has always been the integration layer.

With MCP, that layer is no longer something each productivity vendor has to negotiate one-on-one with each AI vendor. You write one server, you support every client. We've written about how personal AI changes goal achievement, and MCP is the protocol that finally makes that vision routine instead of aspirational.

What can you actually do with the Beyond Time MCP server?

The MCP server exposes your full Beyond Time surface as structured tools. Below is a breakdown by what you'll actually use them for. (A note on terminology: in the Beyond Time UI we use "Goals" and "Milestones." On the server side these correspond to Objectives and Key Results, but you'll only see "Goals" and "Milestones" through the MCP tools, which is the language we'll use here.)

How do you create and manage goals from your AI client?

Goal management uses beyondtime_setup, beyondtime_get_dashboard, beyondtime_list_goals, beyondtime_create_goal, beyondtime_update_goal, and beyondtime_delete_goal.

beyondtime_setup is the orientation call. Run it once at the start of a session and the AI gets a structured view of your account. beyondtime_get_dashboard gives a higher-level snapshot, the equivalent of opening Beyond Time and looking at the home screen. Most assistants call one or both implicitly before doing real work.

The lifecycle is list, create, update, delete. The interesting part isn't the API surface; it's what natural language unlocks on top of it. Try a prompt like:

"Create a Q2 goal called 'Run a half marathon' and add three milestones: build a base of 20 miles per week, complete a 10K race in May, and run 13.1 miles by the end of June."

Claude will call beyondtime_create_goal with the title and timeframe, then beyondtime_create_milestone three times. You don't structure any of it manually. The objects land in your Beyond Time account exactly as if you'd typed them in.

Updating works the same way. "Push the half marathon target date out by four weeks" becomes a beyondtime_update_goal call. "Delete the 'learn French' goal" becomes a beyondtime_delete_goal call after a confirmation.

How do you track milestones and daily actions through chat?

Goals are the destination; milestones are the chunks; actions are what you actually do today. The MCP server exposes all three layers.

For milestones: beyondtime_list_milestones, beyondtime_create_milestone, beyondtime_update_milestone, plus the AI-assisted beyondtime_suggest_milestones. The suggestion tool is the same engine behind the in-app generator. When you ask Claude "break down my new goal into milestones," it can call beyondtime_suggest_milestones, surface a structured plan, and commit chosen ones with beyondtime_create_milestone.

For actions, the surface is broader: beyondtime_list_actions, beyondtime_create_action, beyondtime_update_action, beyondtime_complete_action, beyondtime_schedule_action, and beyondtime_delete_action. Splitting update, complete, and schedule is intentional. Marking done is different from rescheduling, which is different from editing content. Each gets its own tool, so the AI does exactly the right thing.

A typical morning flow:

"What's on my list today? Mark anything from yesterday that I told you I finished as complete, then schedule the deep work blocks for tomorrow morning between 8 and 11."

That message chains beyondtime_list_actions, several beyondtime_complete_action calls, and a few beyondtime_schedule_action calls. The AI orchestrates. You do the work.

How do you build habits and routines via natural language?

Habits and routines are the recurring scaffolding around goals. The server exposes beyondtime_list_habits, beyondtime_create_habit, beyondtime_update_habit, beyondtime_delete_habit, and the matching set for routines: beyondtime_list_routines, beyondtime_create_routine, beyondtime_update_routine, beyondtime_delete_routine.

Two suggestion tools, beyondtime_suggest_habits and beyondtime_suggest_routines, mirror the AI features inside the app. Ask "what habits would move the needle on my half marathon goal?" and Claude can call beyondtime_suggest_habits for a contextual list, then commit chosen ones with beyondtime_create_habit.

Routines differ from habits in that they're sequenced. A morning routine is several blocks in a specific order. The MCP tools let the AI either propose an entire routine in one call or build one up incrementally as you talk through what your real morning looks like.

A useful pattern: ask the AI to read your existing goals, then design a weekly routine that supports all of them. It can chain beyondtime_list_goals, beyondtime_suggest_routines, and beyondtime_create_routine, without you leaving the chat window.

How do you give the AI long-term context about your life?

The trickiest part of any AI assistant is context. The model is brilliant in the moment and amnesiac across sessions. The MCP server addresses this with three personal-context tools: beyondtime_list_personal_contexts, beyondtime_create_personal_context, and beyondtime_delete_personal_context.

Personal contexts are durable notes about you. Things like "I have two kids under five, mornings before 7am aren't realistic for deep work," or "I'm training for a sub-1:45 half marathon, my current 10K is 48 minutes."

These get pulled in by any AI client connecting to your account. When you ask "design a routine for me," the AI doesn't need you to re-explain your life. It calls beyondtime_list_personal_contexts and reads the relevant facts.

Two more delightful tools live here: beyondtime_daily_quote and beyondtime_daily_reflection. They generate context-aware prompts and quotes based on your actual goals. Ask "what's a good reflection question for me tonight?" and you get something rooted in your work, not a generic "what are you grateful for?"

Which MCP clients work with Beyond Time today?

The Beyond Time MCP server implements the Streamable HTTP transport, which is the modern default for remote MCP servers. Any client that supports remote MCP servers over Streamable HTTP can connect.

The clients we've tested most heavily are Claude Desktop and Claude Code, unsurprising given that Anthropic ships and maintains MCP. Both work cleanly: add the server URL, complete the OAuth handshake, and tools light up.

Cursor also supports remote MCP servers and works with Beyond Time. Cursor users tend to use the integration in a slightly different mode, often interleaving goal updates with code work ("I just shipped the auth refactor, mark that milestone complete and add a new one for the post-launch monitoring window").

Beyond those, any spec-compliant client should function. The MCP ecosystem is moving fast and new clients appear every few weeks. Because the Beyond Time server is a standard implementation, we don't do anything specific to support a new client. If it speaks the protocol, it speaks Beyond Time.

If you've evaluated us against a calendar-only competitor like Notion Calendar or a scheduling-first tool like Motion, one quiet but important difference: Beyond Time is built so its data is meant to be touched by AI. Most tools have AI features bolted onto a closed UI. We have an MCP server. That's structural, not marketing.

What does the OAuth flow look like?

Authentication uses OAuth 2.1 with the standard MCP authorization extensions. From your perspective, it's one flow: a browser window pops up, you sign in (or confirm an existing session), you approve, you're back in your AI client with a working server.

Under the hood, the server exposes a WWW-Authenticate header pointing at /.well-known/oauth-protected-resource/mcp, which advertises the OAuth metadata your client needs. Your client handles the rest: browser, code exchange, secure token storage.

Once the token is in place, every MCP request includes it as a bearer token. The Beyond Time server verifies it and resolves to your Supabase session. Every tool call respects the same row-level security policies as the app. The MCP server cannot see data your account isn't already authorized to see.

Tokens are scoped, refreshable, and revocable. Disconnect any AI client at any time from account settings; the moment you revoke, tool calls stop working.

The next question is usually "wait, an AI can write to my goals?" Yes, and the security boundary is identical to you logging in and clicking buttons. The AI acts on your behalf, with your token, against your account. Nothing more.

How is this different from typing into the Beyond Time app?

A fair question. If Beyond Time already has a clean app, why bother with MCP at all? Three reasons.

First, mode-switching cost. The most common user hits goals not at planning time but mid-work. You're writing a strategy doc and realize you need a milestone for customer interviews. You're coding and realize a bug should become a goal-level investigation, not a one-off ticket. With MCP, your AI client (where you're already working) is the input surface. No tab switch.

Second, planning depth. The app is great for fast capture and visualization. But planning where you want to think out loud, get challenged, refine, then commit is better in chat. Talking through a goal with Claude, having it propose milestones, narrowing them, then committing with one tool call is fundamentally different from filling in a form. As our complete guide to personal OKRs argues, quality of structure compounds. AI-assisted planning produces better-structured goals.

Third, automation. Once your goals are addressable as MCP tools, the automation surface expands. Ask Claude to "review my last week, mark anything done, suggest adjustments for next week" as a Sunday ritual. Build agentic flows that pull from your goals and push to your calendar. The app is the primary surface; MCP is the API the AI uses.

Existing workflows aren't going anywhere. The web and mobile apps remain canonical. MCP is additive: another way in, especially for users whose default working environment lives inside an AI client.

What's on the MCP roadmap?

Today's launch covers the full core surface: goals, milestones, actions, habits, routines, personal context, suggestions, daily quotes and reflections, dashboard, and setup. We wanted the first version useful end-to-end, not a half-shipped demo.

A few things next.

Streaming responses on long-running tools. Some operations, particularly suggestions, take a few seconds because they invoke generative models. We'll surface progress more clearly, especially in clients that support streaming.

Resource exposure. MCP supports both tools and resources. Beyond Time leans heavily on tools today. We're scoping which entities make sense as resources (read-only structured data the model pulls in without an explicit call) versus tools (explicit actions).

Webhook subscriptions and proactive nudges. Today the AI acts only when you talk to it. A future version will let your client subscribe to events ("notify me when a milestone falls behind pace") so it can surface things proactively.

Pro-tier MCP features. Some heavier AI tools, deeper history, and bulk operations will move to Pro over time. The base read/write surface will remain on every plan because the connection itself is part of the product.

If you have feedback on what to prioritize, the contact form in the app routes directly to the team. We read everything.

How do you get started in 5 minutes?

Setup is genuinely fast. Here's the sequence.

1. Make sure you have a Beyond Time account. If you don't, sign up at the Beyond Time app. The MCP server uses your Beyond Time account credentials, so you need an account before connecting.

2. Open your AI client's MCP configuration. In Claude Desktop, this is the settings panel under "Connectors" or "Integrations" (UI varies by version). In Claude Code, it's claude mcp add from your terminal. In Cursor, it's the MCP section of settings. Each client has its own UI, but the inputs are the same.

3. Add a new remote MCP server with the URL https://beyondtime.ai/mcp. Don't worry about API keys; the server uses OAuth, not static keys. Save the configuration.

4. Complete the OAuth flow. Your client opens a browser tab. Sign in, approve the connection, and the tab closes itself. Back in your client, Beyond Time tools are now available.

5. Try it. A first prompt that exercises the connection nicely:

"Show me my current goals and tell me which one looks most off-track based on the milestones I've completed."

That single prompt will trigger beyondtime_setup, beyondtime_list_goals, and beyondtime_list_milestones. If you see structured results come back, you're connected. If something goes wrong, the error message will tell you exactly which step failed.

If you're evaluating AI productivity stacks more broadly, our round-up of the best AI productivity apps for 2026 walks through the landscape, including which tools play well together. MCP is increasingly the connective tissue.

Frequently Asked Questions

Is the Beyond Time MCP server free to use?

The MCP server is available to all Beyond Time users on every plan, including the free tier. Some heavier AI-assisted operations (advanced suggestions, bulk operations, longer-history dashboards) will be gated to Pro over time, but the core ability to read and write your goals, milestones, habits, routines, actions, and personal contexts through MCP is part of the base product.

Do I need to know anything about MCP to use it?

No. MCP is a protocol; you don't need to understand the protocol to use it any more than you need to understand HTTP to browse the web. If your AI client supports remote MCP servers (Claude Desktop, Claude Code, Cursor, and a growing list), you paste in the URL, complete an OAuth flow, and use natural language. Everything else is plumbing the AI handles for you.

Can the AI accidentally delete my goals?

Tools like beyondtime_delete_goal exist, so technically yes. In practice, modern clients prompt you before invoking destructive tools, and the AI is cautious without explicit instruction. If you're nervous, work in a read-only mode early ("just list my goals," "just suggest milestones without creating them") until you trust your client's behavior.

Does the MCP server work with mobile or only desktop?

The server itself is platform-agnostic. It works wherever your MCP client works. Today, that's mostly desktop and CLI clients. As mobile MCP clients mature, the same Beyond Time server will work with them. No changes needed on our end.

Will my data be used to train AI models?

No. Beyond Time does not train on your goal data, and the MCP server doesn't change that. When you connect Claude, Cursor, or another client, you're using that client's model to interact with your data. Whether that client trains on prompts is governed by your agreement with that vendor. Beyond Time takes no part in any training pipeline.

What happens if my AI client misuses a tool?

Every tool call is authenticated against your account and respects the same access controls as the app itself. A misused tool can only do things you yourself could do through the UI. If a sequence of calls produces a bad outcome, you can fix it the same way you'd fix any goal-management mistake: open the app, edit, undo. We're also working on richer audit logging so you can review tool calls after the fact.

Can I use the MCP server with my own custom AI agent?

Yes. Any MCP-compatible client works, including custom agents built on Anthropic's or other vendors' SDKs. As long as your agent implements the OAuth flow and Streamable HTTP transport, it connects to https://beyondtime.ai/mcp and calls the same tools Claude Desktop does. Several power users are already building personal automation agents on top of the server.

Where do you go from here with the Beyond Time MCP server?

The Beyond Time MCP server is a small change in surface and a large change in how goals get tracked. The same goals, milestones, and daily actions you'd manage in the app are now reachable from inside the AI client you're already using. The friction between "I should add that to my goals" and "I added that" collapses to a sentence.

If you've been waiting for AI productivity to feel less like a gimmick and more like infrastructure, this is it. The protocol is real. The tools are real. The integration is one OAuth flow away.

Open the Beyond Time app, then point your favorite MCP client at https://beyondtime.ai/mcp. Five minutes of setup, and your goal system starts speaking the same language as your AI.

Related Articles

Tutorials

How to Track Your Goals From Claude (Step-by-Step MCP Setup Guide)

Learn how to track goals from Claude with this step-by-step MCP setup guide. Connect Claude Desktop to Beyond Time and manage goals straight from chat.

Apr 9, 202615 min read

AI

Goal Tracking With MCP: Why Model Context Protocol Is the Future of Personal Productivity

Goal tracking MCP turns your AI assistant into a real coach. See how Model Context Protocol rewires personal productivity around open, agentic workflows.

Mar 19, 202618 min read

Goal Setting

How to Run a Daily OKR Review With Claude (5-Minute Ritual That Actually Sticks)

The 5-minute daily OKR review ritual with Claude and MCP — the morning and evening playbook that finally keeps quarterly goals alive every single day.

Apr 24, 202619 min read

View all articles