Blog
What does LockIn MCP do? How it works with Claude and ChatGPT
LockIn MCP turns your AI assistant into a focus enforcer with system-level hosts blocking, task-aware Chrome filtering, MCP tool calls, and accountable temporary unblocks.
Mil Hoornaert · 10 min read
You tell Claude you need two hours of deep work on a client proposal. It wishes you luck. Twenty minutes later you are on YouTube watching a "quick" tutorial that has nothing to do with the proposal. The assistant had no way to stop you — advice and enforcement are different things.
LockIn MCP bridges that gap. It connects the AI assistant you already use to system-level blocking on your machine, so starting a focus session is as simple as stating what you are working on. This article explains what it actually does, layer by layer, and how it fits alongside tools like Freedom or Cold Turkey.
The problem with standalone blockers
Most distraction blockers live in a separate app. You open it, configure a blocklist, set a schedule, and hope you remember to activate it before the hard part of your task begins.
That separation is the failure mode. When work gets difficult, opening a second app to manage focus is friction you skip — and a browser extension you can disable from the toolbar takes about three seconds to bypass.
LockIn MCP instead connects to Claude, ChatGPT, Perplexity, Gemini, and Poke through the Model Context Protocol (MCP). The assistant you are already talking to becomes the control surface for enforcement. For a full comparison of where this sits among other tools, see best distraction blockers in 2026.
How it works, layer by layer
LockIn MCP is not a single trick. It stacks four enforcement layers that close different bypass routes.
Layer 1: System-level domain blocking
When a session starts, LockIn edits your system's hosts file to redirect distracting domains — commonly Reddit, YouTube, and X, all customizable — to a dead address (127.0.0.1). This operates below the browser layer, so every browser on your machine is blocked at once. Switching from Chrome to Firefox does not help.
Undoing a hosts-file block requires terminal access and admin permissions — real friction compared to clicking "disable extension." Learn more in what is a hosts file and system-level blockers for 2026.
Layer 2: Chrome extension with task-aware filtering
A static blocklist cannot tell the difference between YouTube for a work tutorial and YouTube for passive scrolling. The LockIn Chrome extension evaluates whether the page you are viewing relates to your currently stated focus task.
This catches edge cases a blanket domain block would over-block or under-block. Domain-level blocking stops the impulse visit; task-aware filtering handles the gray area.
Layer 3: MCP server connecting your AI assistant
This is the piece that separates LockIn from every standalone blocker. Through MCP, your connected assistant can call tools that:
- Start and end focus sessions
- Check your active task
- Enforce blocking
- Manage temporary unblocks
You do not switch to a settings panel mid-task. You tell the assistant what you need, and it acts. That is the core of AI productivity distraction blocking.
Layer 4: Temporary unblocking with accountability
Legitimate exceptions happen. You need a reference doc on a blocked domain, or a ten-minute tutorial video. Instead of silently disabling the block, you ask your assistant for a time-boxed unblock.
The exception is logged, time-limited, and closes automatically when the timer runs out. The friction of asking — rather than clicking a toolbar toggle — interrupts the impulsive bypass loop. Research on interruption recovery suggests why that matters: Gloria Mark's work at UC Irvine, widely cited as the 23-minute rule, found that regaining deep focus after an unplanned interruption can take more than 20 minutes. Reducing how often you start that recovery clock is the actual goal.
A practical example: two-hour proposal session
- You tell Claude: "I'm working on the Acme client proposal for the next two hours. Block my usual distractions."
- Claude calls LockIn's MCP tools. The hosts-file block activates for your configured list.
- Twenty minutes in, you catch yourself typing
youtube.com. It does not load. - You genuinely need a 10-minute tutorial on slide formatting. You ask Claude for a temporary unblock with a 10-minute window.
- Claude grants it, logs it on your dashboard, and access closes automatically when the timer ends.
- You close the Claude window. The background service keeps blocking and timers running.
No separate app. No schedule you configured yesterday that does not match today's actual work.
Why the AI layer matters
Two things change when enforcement runs through your assistant:
Reduced activation friction.You do not need to remember a separate tool. The conversation where you plan your work is the same conversation that starts protection.
Decision friction at the bypass moment.Asking for an exception out loud (or in chat) is harder than clicking "pause blocker." That small gap is often enough to break an automatic habit loop — the kind where your hands type reddit.com before your conscious mind catches up.
The American Psychological Association's research on multitasking switching costs aligns with what knowledge workers report: the cost is not the two minutes on the distracting site. It is the recovery afterward. Blockers help by adding friction at the start of that switch, not by removing willpower entirely.
How LockIn MCP compares to familiar tools
| Dimension | Standalone blocker (Freedom, Cold Turkey) | LockIn MCP |
|---|---|---|
| Session trigger | Manual schedule or app UI | Natural language via AI assistant |
| Knows current task | No — fixed blocklist | Yes — stated in conversation |
| Enforcement layer | Varies (app-level or system) | Hosts file + Chrome extension |
| Unblock model | Settings toggle or wait-out timer | Time-boxed, logged, AI-requested |
| AI integration | None | Native MCP (Claude, ChatGPT, etc.) |
For head-to-head comparisons, see LockIn vs Freedom and LockIn vs Cold Turkey.
Dashboard visibility without surveillance
LockIn logs distraction attempts during sessions — when you tried to reach a blocked domain and whether you requested an unblock. It does not expose your full browsing history.
The dashboard is built for accountability: seeing that you hit Reddit four times during a writing block is useful signal about what to tighten next. It is not built for surveillance of everything you read online.
Getting started
Installation is one command:
npx -y lockin-mcp installWorks on macOS, Linux, and Windows. Full setup steps are in how to install LockIn MCP. Connector details per assistant are in docs/connectors.
Pricing: 3-day free trial, then $2.99/month or $59.99 lifetime.
Frequently asked questions
Do I need a specific AI assistant?
No. LockIn supports Claude, ChatGPT, Perplexity, Gemini, and Poke via MCP. Use whichever you already work in.
Does LockIn MCP replace my browser extension blocker?
It can. System-level hosts blocking covers every browser at once, and the Chrome extension adds task-aware filtering on top. Many users uninstall standalone extension blockers after connecting LockIn.
What happens if I close my AI chat window mid-session?
The background service keeps the MCP server and active unblock timers running independently of the chat window. Blocking persists until the session ends or the timer expires.
Can I use LockIn alongside Freedom or Cold Turkey?
Some people use Freedom for mobile scheduling and LockIn for AI-assisted desktop deep work. They solve overlapping but not identical problems. See Freedom vs AI blocker for when each fits.
How much does it cost?
3-day free trial, then $2.99/month or a one-time $59.99 lifetime option. Details at /get-started.
Is my data sent to LockIn's servers?
Blocking runs locally on your machine. Session metadata and distraction attempt logs sync to your dashboard; your full browsing history does not. See how it works for the architecture overview.
