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

This guide shows you how to track goals from Claude using the Model Context Protocol (MCP). By the end, you will be able to talk to Claude in plain English and watch your Beyond Time goals, milestones, habits, and routines update in real time. No copy-paste between apps. No screen-switching. Just a conversation that quietly maintains your goal system in the background.

If you already use Claude as your daily thinking partner, this is the most direct way to keep your goals current without breaking flow. You ask Claude to plan your week, and the same conversation creates milestones. You tell it what you did today, and your habits get logged.

This walkthrough covers the credentials you need, the server config for Claude Desktop, the OAuth handshake, the first prompts to verify the connection, and a real workflow: creating a goal, breaking it into milestones, attaching habits and routines, and running a weekly review. We end with troubleshooting and an FAQ.

What do you need before you start?

Before you wire anything up, line up four prerequisites. Skipping ahead and debugging in reverse is what eats most of the time on these setups.

1. A Beyond Time account. You need an active account at beyondtime.ai. The free tier is enough to test the connection.

2. Claude Desktop or another MCP-capable client. This guide focuses on Claude Desktop. The same endpoint works with Claude Code, Cursor, and any client that speaks Streamable HTTP MCP.

3. A modern Node runtime (some clients only). Claude Desktop ships its own MCP runtime. Claude Code uses your shell; most setups have Node 18+.

4. About 10 minutes for OAuth. The first connect opens a browser, you sign in to Beyond Time, and a token is exchanged. After that, Claude reconnects silently.

The MCP endpoint you will be pointing clients at is:

https://beyondtime.ai/mcp

That is the only URL you need. There is no per-user subdomain or API key. Authentication is handled by OAuth at connection time.

For background on why this matters, our piece on AI-augmented goal achievement explains the shift from "AI as a feature" to AI as the connective tissue of your planning system. MCP is the protocol that makes it possible.

How do you connect Claude Desktop to the Beyond Time MCP server?

Connecting Claude Desktop is a three-step dance: edit a config file, restart the app, complete an OAuth handshake. Each step has a specific failure mode, so we will walk through them carefully.

How do you add the server config to Claude Desktop?

Claude Desktop reads MCP servers from a JSON config file. The location depends on your operating system.

On macOS, the file lives at:

~/Library/Application Support/Claude/claude_desktop_config.json

On Windows, it is at:

%APPDATA%\Claude\claude_desktop_config.json

If the file does not exist yet, create it. If it does exist and you already have other MCP servers configured, you will be adding a new entry inside the existing mcpServers object.

The config block for Beyond Time looks like this:

{
  "mcpServers": {
    "beyondtime": {
      "url": "https://beyondtime.ai/mcp"
    }
  }
}

That is the entire configuration. No command, no args, no environment variables. The Beyond Time MCP server is hosted, not local, so you are pointing Claude at a remote URL. Claude Desktop's built-in MCP client handles the rest.

If you already have entries like filesystem, github, or other MCP servers, just add beyondtime as another key alongside them. Make sure the JSON stays valid. A trailing comma or a missing brace is the most common reason this step fails silently.

Save the file. Then quit Claude Desktop completely (not just close the window) and reopen it. On macOS that means Cmd+Q. On Windows, right-click the tray icon and choose Quit. Reopening is what triggers Claude to read the new config.

How does the OAuth handshake work?

The first time Claude Desktop tries to call a Beyond Time tool, it discovers the server requires authentication. Specifically, our endpoint returns a WWW-Authenticate header pointing at the OAuth resource metadata, which Claude reads to find the authorization server.

Claude opens your default browser to a Beyond Time consent screen. You sign in, confirm that you want to grant Claude access to your goals data, and the browser hands the token back via a localhost callback.

The token is scoped to your account. Claude can read and write your goals, milestones, habits, routines, actions, and personal context, plus call AI suggestion tools. It cannot see other users' data. Every query passes through Supabase Row Level Security.

Tokens are long-lived but refreshable. If your session expires, you will see an "authentication required" error and Claude will prompt you to re-auth.

How do you confirm the connection works?

The fastest way to confirm is to ask Claude to list your goals. Open a new conversation in Claude Desktop and type:

"List my Beyond Time goals."

Claude should call beyondtime_list_goals and return whatever you have. If your account is empty, you will see a message like "No goals found. Create one with beyondtime_create_goal." That is success, not failure. The tool was called, the auth worked, and the response came back.

If Claude says it cannot find any Beyond Time tools, the config did not load. Quit and relaunch Claude Desktop. If you see an OAuth error, clear cookies for beyondtime.ai and retry.

For a richer first signal, run:

"Show me my Beyond Time dashboard summary."

That triggers beyondtime_get_dashboard, which returns a single payload covering your goals, top milestones, today's habits, and pending actions. It is the most efficient way to orient Claude inside your data.

How do you connect Claude Code or other MCP clients?

If you live in the terminal, Claude Code supports the same MCP server with a slightly different config path. Claude Code reads from ~/.config/claude-code/config.json (or the platform equivalent), and the structure is identical to Claude Desktop's.

{
  "mcpServers": {
    "beyondtime": {
      "url": "https://beyondtime.ai/mcp"
    }
  }
}

Restart Claude Code. The first tool call triggers OAuth in your default browser. If you are SSH'd into a remote machine, the flow gets trickier; Beyond Time's MCP works best with clients that have local browser access.

For Cursor, use the Settings UI under "MCP Servers" and add https://beyondtime.ai/mcp as a new HTTP server. Other clients that support Streamable HTTP MCP should all work the same way.

If you are evaluating tools for this kind of integration, our roundup of the best AI productivity apps for 2026 covers which products actually expose MCP today.

What's the first prompt you should run to confirm everything works?

The very first prompt should not create or modify anything. Use a read-only tool to make sure auth, transport, and parsing are all clean before you write data.

The cleanest first prompt is:

"Pull my Beyond Time dashboard and tell me what stands out."

Claude calls beyondtime_get_dashboard, gets a JSON payload of your goals, milestones, habits, routines, and recent actions, and summarizes it back. You get two confirmations: the connection works and the model can reason over the data.

If your account is empty, follow up with:

"Set up Beyond Time for me with a starter goal: 'Run a half marathon by October.' Use beyondtime_setup if it helps."

beyondtime_setup initializes a new user with sensible defaults. Skip it if your account is already populated.

For an existing account, try:

"Pull my goals, find the one about fitness, and suggest 5 milestones using beyondtime_suggest_milestones."

This confirms reads, AI suggestions, and chained calls in one shot.

How do you create your first goal through Claude?

Once reads work, writes are straightforward. You do not have to memorize tool names; Claude picks them based on intent. But knowing what is happening under the hood helps you write better prompts.

When you say "create a goal about learning Spanish," Claude calls beyondtime_create_goal with a title argument. The server returns the new ID, which Claude reuses for follow-up calls in the same conversation. That is how chained operations work.

A good first prompt:

"Create a Beyond Time goal called 'Ship Beyond Time MCP integration to 1,000 users by end of Q3.'"

Claude calls beyondtime_create_goal and confirms the new ID. If you follow up with "Now break it into 5 milestones," Claude chains beyondtime_suggest_milestones and beyondtime_create_milestone automatically.

What to avoid: vague prompts. "Help me plan something" leaves Claude guessing between goals, milestones, and routines. "I want a goal to read 24 books this year" gives it exactly what it needs.

The same logic behind breaking down big goals into actionable steps applies here, except the notebook writes itself.

How do you add milestones, habits, and routines through chat?

Goals are the top of the hierarchy. Milestones are measurable checkpoints. Habits are recurring inputs. Routines are time-boxed sequences. Claude can move between all four in one conversation.

How do you add milestones in chat?

For milestones, the underlying tool is beyondtime_create_milestone, which takes a goal_id, a title, and an optional target_date. If you already have a goal in the conversation context, Claude reuses its ID. If not, it lists goals first, then creates against the right one.

A working prompt:

"For my 'Ship MCP integration' goal, add three milestones: 'Publish setup guide' by April 30, 'Reach 250 connected accounts' by June 15, and 'Hit 1,000 connected accounts' by September 30. Use beyondtime_create_milestone for each."

That triggers three sequential beyondtime_create_milestone calls. Each one returns an ID, and Claude reports back the full list with target dates so you can scan for typos. If you want AI to do the breakdown for you, say "suggest milestones for this goal" and Claude calls beyondtime_suggest_milestones, which uses the same OKR-aware engine described in our complete guide to personal OKRs.

How do you add habits and routines in chat?

Habits use beyondtime_create_habit. Routines use beyondtime_create_routine. Both link back to goals, so you do not lose the connection between daily inputs and long-term outputs. This is the same principle behind our argument that habits need to be connected to goals: isolated habits drift, but habits anchored to goals compound.

A representative prompt:

"Add a daily habit 'Write 500 words on the MCP setup post' linked to my MCP goal, and create a morning routine called 'Builder block' that includes 90 minutes of deep work. Use beyondtime_create_habit and beyondtime_create_routine."

Claude executes both, returns the IDs, and now your dashboard shows the new habit and routine without you opening the app. If you want suggestions instead of dictating, ask "suggest habits and routines for my MCP goal" and Claude calls beyondtime_suggest_habits and beyondtime_suggest_routines.

How do you log actions on the fly?

For day-of execution, beyondtime_create_action and beyondtime_complete_action let you treat Claude like a capture inbox. "Add an action: review draft, due today" creates it. "Mark the draft review as done" completes it. Use beyondtime_schedule_action for a specific time block.

This is where the integration earns its keep: the gap between thinking about a task and recording it collapses to one sentence.

How do you do a weekly review through Claude?

Weekly reviews are the single highest-leverage habit in any goal system, and they are also the one most people skip. The reason is friction: opening the app, reviewing each goal, updating progress, deciding what is next. With Claude in the loop, the whole review can happen as one focused conversation.

Here is a battle-tested weekly review prompt:

"Run my Beyond Time weekly review. Pull the dashboard, list goals and their milestones, summarize what moved this week based on completed actions, then suggest progress updates for each milestone. Use beyondtime_get_dashboard, beyondtime_list_goals, beyondtime_list_milestones, and beyondtime_list_actions. Once we agree on updates, call beyondtime_update_milestone for each."

Claude works top-down: pulls a dashboard, inventories goals and milestones, scans completed actions to infer what moved, then proposes progress percentages. You confirm, and beyondtime_update_milestone persists the new numbers.

The conversation usually takes 5 to 8 minutes versus 20 to 30 manually. More importantly, it actually happens.

To make the review reflective, add:

"Pull a daily reflection prompt with beyondtime_daily_reflection and walk me through it. Save anything important to personal context with beyondtime_create_personal_context."

beyondtime_daily_reflection returns a prompt grounded in real progress data. beyondtime_create_personal_context saves the insight as long-term memory the system can reference later, so suggestions get more tailored over weeks.

If the review surfaces dead weight, beyondtime_delete_goal, beyondtime_delete_habit, beyondtime_delete_routine, and beyondtime_delete_personal_context are all available. Pruning matters; our piece on tracking multiple goals without overwhelm argues ruthless pruning is what keeps a goal system alive.

What happens when something goes wrong?

Things will go wrong. The first OAuth attempt times out, the config has a typo, a tool returns an error you do not understand. Here is the short troubleshooting playbook.

"Claude doesn't see any Beyond Time tools." The config did not load. Validate the JSON, confirm the file path for your OS, and fully quit and relaunch Claude Desktop. Reopening the window is not enough.

"OAuth fails or hangs." Confirm you can reach https://beyondtime.ai in a normal browser. Some corporate networks block unknown OAuth callbacks. If the consent screen loads but redirect fails, check for a firewall or extension blocking localhost.

"Authentication expired." Run any tool, get the 401, follow Claude's prompt to re-auth. Takes 30 seconds.

"Specific tool error." Read the message. "Goal ID not found" means the ID was deleted. "Validation failed" usually means a title was too long. Ask Claude to retry with corrected input.

"It worked yesterday and now it doesn't." Check Beyond Time's status. Restart Claude Desktop. Sign out, sign back in, re-run OAuth.

"Claude is being weird about which goal to update." Be explicit. "List my goals first, then update the one with ID X." When Claude has the ID in context, it stops guessing.

For the philosophy behind treating goals as a connected system, our scattered goals to focused action writeup is a useful companion read.

What other questions come up about tracking goals from Claude?

Do I need to be on Beyond Time Pro to use MCP?

No. The MCP server is part of the core Beyond Time platform and works with any active account. Pro features (deeper personal context, unlimited goals, priority AI tools) layer on top of the same MCP surface. If you want to see what Pro includes, our Pro version page breaks it down.

Is the OAuth token safe? What can Claude actually do with it?

The token is scoped to your account and to the tools registered in the MCP server. Claude can read and write your goals, milestones, habits, routines, actions, and personal context, plus call AI suggestion tools. It cannot access other users' data, billing, or anything outside the published tool list. Supabase Row Level Security enforces the boundary at the database layer.

Can I use this with ChatGPT or other non-Claude assistants?

Anything that speaks the official MCP Streamable HTTP transport works. That currently includes Claude Desktop, Claude Code, Cursor, and a growing list. ChatGPT does not natively support MCP yet.

What if I have multiple Beyond Time accounts?

Each MCP connection is tied to one OAuth session. To run two accounts in parallel, use two separate clients (e.g., Claude Desktop and Claude Code), each authenticated separately.

How is this different from just asking Claude for goal advice without MCP?

Without MCP, Claude gives advice but cannot touch your data. You copy suggestions in by hand. With MCP, suggestions become real records in your account in the same turn. The friction reduction is the entire point.

Will the tool list change over time?

Yes, slowly. The current set covers the full goal management lifecycle. Ask Claude "what Beyond Time tools do you have available?" to see the live list.

Does this work offline?

No. The MCP server is hosted at https://beyondtime.ai/mcp, so you need an internet connection.

Why is now a good time to track goals from Claude?

A year ago, you could ask an LLM for goal advice but could not connect it to anything real. Now you can talk to Claude about your week and watch your dashboard update in the same minute. That is the gap MCP closes.

To track goals from Claude well, you do not need to memorize 30 tool names. You need a clean account, a working MCP connection, and a few habits: pull the dashboard before planning, create goals in plain English, ask for milestone suggestions when stuck, and run a weekly review through chat. The tools handle the rest.

The setup is short. The payoff compounds because every conversation keeps your goal system slightly more current. That currency is what most goal systems lose to neglect, and what this integration preserves.

If you have not connected yet, open Beyond Time, sign in, add the MCP config to Claude Desktop, and ask Claude to pull your dashboard. Five minutes from now you will have a working integration.

Related Articles

Tutorials

The Complete Beyond Time MCP Tools Reference: Every Tool, Every Prompt

The canonical Beyond Time MCP tools reference: every tool, every parameter, every example prompt. Bookmark this developer-grade guide and ship faster.

Apr 17, 202618 min read

Product

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

The Beyond Time MCP server is live. Connect Claude, Cursor, and other AI clients to your goals, milestones, habits, and routines through natural language.

Mar 4, 202617 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