Agent-inbox is an open-source MCP tool for feeding links from phone, browser, or CLI to Claude and ChatGPT. It creates persistent, agent-driven reading lists.
You are reading an article on your phone, and you want your AI assistant to analyze it later. In most workflows, that important link stays stranded in a browser tab or buried in a notes app. Agent-inbox solves this by giving AI agents a real inbox where you can deposit links from a phone, a browser, or a command-line interface. This open-source project, created by OGZamasu, connects directly to MCP-compatible clients such as Claude and ChatGPT. Once a link lands in the inbox, the agent can retrieve it and respond with an opinion and suggested next steps. This turns a simple bookmarking habit into an asynchronous, agent-driven reading workflow.
Agent-inbox is a JavaScript-based open-source project designed to solve a practical problem: how to get web content from any device into an AI conversation. It provides a persistent inbox that can accept links from multiple surfaces, including a Chrome extension, a CLI tool, and a mobile-friendly web flow. The backend runs on Cloudflare Workers, so there is no dedicated server to manage.
The project is built around Model Context Protocol (MCP), an emerging standard that lets AI assistants connect to external tools and data sources. Because agent-inbox uses MCP, it works with any MCP-compatible client. For most users, that means Claude and ChatGPT today, but the same architecture can extend to other assistants as they adopt MCP.
The typical AI chat workflow is session-based. You open a conversation, paste a link or question, receive an answer, and then move on. That model works well when you are focused, but it fails when you are away from your desk or when you are collecting information over time.
Most people discover valuable content on their phone, in a browser, or in a newsletter. In a traditional workflow, that content either disappears or requires manual copy-pasting later. Agent-inbox closes that gap by giving you a persistent, cross-device capture mechanism. You can deposit links throughout the day and let the agent process them whenever you are ready.
For developers and power users who rely on assistant tools like Claude and ChatGPT, this is a significant improvement. It aligns with the growing interest in asynchronous AI agent workflows, where users queue inputs for agents to process later. Instead of forcing every interaction to happen in real time, agent-inbox lets you build an agent-driven reading list that is always available.
MCP is the key technical enabler for agent-inbox. In simple terms, MCP provides a standard way for AI models to call tools, read resources, and access data. Agent-inbox exposes a link inbox as an MCP resource, allowing compatible clients to query it directly.
This design avoids the need for custom integrations for every assistant. As long as a client supports MCP, it can read from the inbox. That is why the project can support both Claude and ChatGPT without maintaining separate connectors for each platform.
The MCP alignment also reflects a broader trend in the AI ecosystem. Since 2024, MCP has emerged as a common interoperability layer between AI applications and external systems. Tools that adopt MCP early can benefit from a growing ecosystem of clients and servers.
The workflow for agent-inbox is straightforward:
This flow makes it easy to move from passive consumption to active analysis. You do not need to keep a conversation open or remember to copy links into a prompt later. The inbox holds them until you are ready.
Agent-inbox is useful for anyone who wants to offload content processing to an AI assistant. Here are several ways to apply it in daily work:
These use cases share a common pattern: capture content in context, process it later, and let the agent do the heavy lifting. For busy developers, this reduces the friction between discovery and understanding.
The agent-inbox stack is deliberately JavaScript-based and lightweight. The serverless backend runs on Cloudflare Workers, which provides global distribution, automatic scaling, and a simple deployment model. Developers who are already familiar with Cloudflare tooling can get started quickly.
The Chrome extension and CLI serve as input surfaces for capturing links. The extension is ideal for desktop browsing, while the CLI works well for developers who live in the terminal. The project also accommodates phone-based capture through a web-friendly flow, making it a truly cross-device solution.
Using Cloudflare Workers as the backend has several benefits for an open-source project:
For a tool designed to handle links and support a small community, this stack keeps the project approachable and affordable to run.
Agent-inbox sits at the intersection of three important trends shaping AI tooling today.
MCP ecosystem growth. The Model Context Protocol has gained significant momentum since 2024. More AI clients and tools are adopting MCP as a standard way to communicate. Agent-inbox is positioned to ride this wave by using MCP natively.
Asynchronous AI agent workflows. Users are moving away from one-shot chat interactions and toward workflows where agents work through queues and tasks over time. A persistent inbox is a natural component of that model.
Browser extensions and CLIs as AI input surfaces. As AI assistants become more embedded in daily routines, the ways users capture input are diversifying. Agent-inbox combines a browser extension and a CLI to meet users where they already work.
These trends point to a future where AI is not just a chat window, but an automated partner that can be fed from many sources and can respond when appropriate.
If you use Claude, ChatGPT, or another MCP-compatible client, agent-inbox is easy to add to your workflow. Start by exploring the open-source repository to understand the setup and deployment options.
The general setup pattern is straightforward:
Because the project is open source, you can also customize it to fit your own needs. If you discover a missing feature, you can contribute code, open an issue, or suggest an improvement. The project is part of a larger effort to make AI workflows more persistent and interoperable.
Agent-inbox is a practical answer to a common problem: useful content is discovered everywhere, but it rarely makes it into the AI assistant on your desk. By providing a persistent inbox with MCP-compatible access, the project helps developers and power users turn bookmarks into actionable intelligence.
The tool is easy to try, built on a modern serverless stack, and aligned with the most important trends in AI tooling. Whether you want to build a reading list, track competitors, or simply avoid losing great links, agent-inbox offers a clean path forward. Start by experimenting with the open-source project and connecting it to Claude or ChatGPT. The next time you see something worth reading, you will have a place to send it and an agent ready to respond.
Agent-inbox is an open-source, JavaScript-based tool that gives AI agents like Claude and ChatGPT a persistent inbox for web links. You can send links to it from your phone, browser, or command line, and the AI agent can later retrieve them for analysis, allowing for asynchronous, agent-driven reading workflows.
Agent-inbox provides multiple capture surfaces: a Chrome extension, a command-line interface (CLI), and a mobile-friendly web flow. You choose whichever is most convenient for the device you're using—click the extension in your browser, use the CLI from a terminal, or open the web flow on your phone—and the link is immediately stored in your inbox.
Yes, agent-inbox is built around the Model Context Protocol (MCP), an emerging standard for connecting AI assistants to external tools. Because it is MCP-compatible, it works with any MCP-compatible client, which currently includes both Claude and ChatGPT. It can also extend to other assistants as they adopt MCP.
The project is designed to run on Cloudflare Workers, so you don't need to set up or maintain a dedicated server. Since it's open-source, you can deploy it to your own Cloudflare account, but there's no complex backend to manage. The core requirement is having a Cloudflare Workers environment for the backend.
Regular bookmarking tools simply store links for you to read later. Agent-inbox connects those stored links directly to AI agents like Claude and ChatGPT, allowing the agent to retrieve the content, form an opinion, and suggest next steps. This transforms a passive list into an active, agent-driven workflow where your AI assistant can process your reading queue whenever you're ready.
