Instapaper MCP: Connect Your Reading List to AI (2026)

You saved the article three weeks ago. You meant to read it. It is still sitting in Instapaper with forty others just like it, and you could not tell me the title of a single one. The read-later pile is where good intentions go to wait, and the honest truth is that most of what lands there never gets read at all.

That pile is not useless, though. It is a hand-picked list of things you already decided were worth your time. The problem was never the saving. It was that saving and reading are two different jobs, and only one of them ever actually happens. So here is a better question. What if the reading did not have to be you?

That is what an Instapaper MCP server does. It hands your AI assistant a live connection to your Instapaper account, so Claude can read, search, summarize, and organize the pile on your behalf. Ask “what did I save about pricing strategy last month?” and the agent pulls the articles, reads them, and gives you the answer. This guide covers what it does, the fact that there is no official server and what to do instead, the setup, and the honest limits.

What Instapaper MCP actually does

The Model Context Protocol is an open standard Anthropic released in late 2024. It defines how an AI client talks to an outside tool. A server exposes a set of tools, the client calls them mid-conversation, and the results flow straight back into the model’s reasoning. No copy-paste, no switching apps, no leaving the chat.

Instapaper is a read-later app. You save an article from any browser or app, it strips out the ads and clutter, and it syncs a clean copy to every device. After Pocket shut down in July 2025, a lot of people landed on Instapaper, and Kobo picked it to replace Pocket on its e-readers. It is one of the oldest and most stable read-later services still running.

An Instapaper MCP server wires that account to your AI client. Once connected, the assistant can list your saves, read the full text of any article, search across everything, star and archive, move things between folders, and pull your highlights. The community server exposes more than 30 tools across five groups, covering content management, bulk operations, folders, highlights, and search. In practice that means the agent can do anything you would do in the app, only faster and in bulk.

The shift here is small but real. Instapaper already saves and cleans articles beautifully. What it does not do is think about them for you. The MCP server closes that gap. Your saved pile stops being a graveyard of links and becomes a live corpus the agent can read on command.

The catch: there is no official server

Search “instapaper mcp” and here is the thing nobody says up front. Instapaper does not publish its own MCP server. It ships a documented, stable public HTTP API, and that is what everything below is built on, but there is no first-party connector you can just switch on the way you can with Notion MCP. That leaves you two honest routes.

The community server (the capable one): The open-source hendronf/Instapaper-MCP is the real deal, with 30+ tools and production-ready releases. It runs locally on your machine. You clone it, build it, and give it your Instapaper API credentials. This is the pick if you want the full tool surface and you are comfortable running a node process.

The no-code bridge (the easy one): Platforms like Zapier MCP, viasocket, and Glama can wrap Instapaper’s API and hand you a hosted MCP endpoint. No install, no build step, you connect Instapaper through the platform’s own login. The trade is a thinner tool set and a third party sitting in the middle of your data.

The choice is not really about convenience. Both work. The decision is trust. The community server runs on your own hardware but wants your Instapaper login, and you are trusting a solo developer’s code with it. The hosted route hides the credentials but routes your reading history through a SaaS you now have to trust instead. Neither is official, so pick the one whose trust model you can live with, and read the section on credentials below before you commit. The rest of this guide uses the community server, because it is the one with the tools worth having.

How to connect Instapaper MCP to your AI client

The community server is a local process, so setup looks different from a hosted URL server. Three steps, then a client config.

Step 1: Get API keys. Request access at instapaper.com/api. Instapaper emails you a consumer key and secret. Approval is manual and can take a few days, so start here first. The server also needs your Instapaper email and password, because the API uses OAuth 1.0 to trade those for an access token.

Step 2: Build the server. You need Node 18 or newer. Clone the repo, run npm install, copy .env.example to .env, fill in your four values, and run npm run build. That produces the file your client will launch.

Step 3: Add it to your client. Point your AI client at the built server. Here is the Claude Desktop config, which is the most common setup.

{
  "mcpServers": {
    "instapaper": {
      "command": "node",
      "args": ["/absolute/path/to/instapaper-mcp/build/index.js"],
      "env": {
        "INSTAPAPER_CONSUMER_KEY": "your_consumer_key",
        "INSTAPAPER_CONSUMER_SECRET": "your_consumer_secret",
        "INSTAPAPER_USERNAME": "your_email",
        "INSTAPAPER_PASSWORD": "your_password"
      }
    }
  }
}

Whichever client you use, fully restart it after editing the config. A window reload does not pick up a new server, and that one missed step is behind most “my server is not working” complaints. Then ask “list my ten most recent Instapaper saves” to confirm the tools loaded.

What you can actually do with it once connected

Reading tools sound obvious until you have used one on a real pile for a week. Here is the short list people actually reach for.

Triage the backlog: “Summarize everything I saved this week in two lines each, and tell me which one is worth reading in full.” The agent reads the pile and hands you a menu, so you spend your reading time on the one article that earns it instead of scrolling a list of titles you no longer remember saving.

Search by meaning, not just words: Instapaper’s own full-text search is a Premium feature. An agent works around that. “Find the articles I saved about cold email that mention deliverability” makes the model fetch and read, so you get a meaning-based search over your saves even on the free tier.

Turn saves into output: “Draft a short brief from the five articles I starred about onboarding, with the three ideas they agree on.” Your read-later pile stops being an input you never process and becomes raw material the agent shapes into something you can use.

Do the housekeeping you keep avoiding: “Archive everything older than six months that I never opened.” Bulk operations across hundreds of saves in one instruction, the kind of cleanup you would never do by hand.

“What’s the shortest unread article I saved? Give me the one I can finish in three minutes.”

That single prompt is the whole pitch in miniature. Not another place to save a link, but a way to finally get value out of the links you already saved.

Where Instapaper MCP still needs a human

This server reads your account, but it also writes to it. That changes the risk profile, so the honest limits are worth reading before you lean on it.

You are handing over your credentials: The community server needs your Instapaper email and password to build its access token, and it is code written by a third party running on your machine. That is not a reason to avoid it. It is a reason to read the repo, check the release history, and treat those credentials the way you would any password you paste into a tool you did not write. For the broader version of this question, see is MCP safe.

It can archive and delete, not just read: The same tools that tidy your pile can also move or remove things you wanted. Keep tool approval on, so the agent asks before a bulk archive wipes a folder you meant to keep. Read-only would be safer, but this server is built to manage, not just fetch.

It only knows Instapaper: This is the big one. The server reaches your Instapaper saves and nothing else. The long-form articles you filed to read later are only one shelf of what you actually keep. The rest is scattered, which is the whole point of the next section.

Free-tier ceilings still apply: Instapaper Premium at $5.99 a month unlocks full-text search and unlimited notes, where the free tier caps notes at five a month. The agent routes around search by reading directly, but if you rely on notes or the permanent archive, the tier matters.

The mistakes that break Instapaper MCP setups

The same handful of issues come up over and over.

Not waiting for API approval: The consumer key and secret come from a manual review that takes days. People try to wire it up the same afternoon, get an auth error, and blame the server. Request the keys first, build everything else while you wait.

Reloading instead of restarting: Claude Desktop reads its server list at full startup only. Reload Window does nothing. Quit fully, reopen, then check for the tools.

Pasting credentials without reading the code: This is a community server, not an official one. Skim the source, check the releases, and know what you are running before you drop your password into the env block. Ten minutes of reading is cheap insurance.

Expecting it to reach your other saves: It reads Instapaper. It has no idea about the thread you bookmarked on X or the comment you saved on Reddit. When the agent says it cannot find something, it is often looking in the one shelf it can see.

Running too many servers at once: Each MCP server is a subprocess with startup and memory cost. Past ten, you feel it. Three to six is the sweet spot. For which ones earn a slot, see the best MCP servers for knowledge workers.

Why your reading list is only half your saved knowledge

Here is the gap the Instapaper setup leaves wide open. Instapaper holds the long reads. The articles, the essays, the deep-dive you filed to get to on the weekend. That is real, and it is worth connecting. But it is not where most of what you save actually lives.

Think about the last week. The prompt you saved was an X thread. The workaround that finally fixed your build was a Reddit comment. The framework you keep meaning to try was a LinkedIn post. None of that goes to Instapaper. It sits in three separate in-app bookmark tabs, unsearchable, invisible to any agent, slowly disappearing under everything you saved after it. Instapaper MCP reads one shelf. The other three stay dark.

So you can give Claude your reading list and still have an assistant that cannot recall the single most useful thing you saved, because you saved it on a social platform instead of a read-later app. Connecting Instapaper is a real upgrade. It is also a partial one.

This is where a bookmarks MCP earns its slot next to it. ContextBolt is a Chrome extension that captures what you save on X, Reddit, and LinkedIn, tags each save by topic automatically, and exposes the whole collection through a personal MCP endpoint. The free Basic tier covers 150 bookmarks. Pro at $6 a month adds unlimited saves, encrypted cloud sync, and the MCP endpoint any client can call. Wire it in the same way you wired Instapaper. Now the agent reads both halves in one prompt. “Pull anything I saved about onboarding” returns the Instapaper article and the two X threads side by side. For the bigger picture of feeding an agent your own context, see the personal AI context stack, and if you are here because Pocket shut down, the Pocket alternatives roundup covers where the long reads should go.

The one opinion worth holding

Most people will set up Instapaper MCP to save articles faster, and that is the smallest possible use of it. Saving was never your problem. You are already brilliant at saving. The pile proves it.

The real unlock is triage. The moment an agent can read the whole backlog and tell you which two articles out of forty are worth an hour, the read-later list stops being a source of guilt and starts being a resource. That is the shift worth making. Not one more thing that captures links, but the first thing that finally reads them back to you.

And once you have felt that, the limit becomes obvious. An assistant that reads your articles but not your saved threads knows half your mind. Connect the reading list, then connect the rest, and the agent finally works from everything you decided was worth keeping, wherever you happened to keep it. That is the whole setup. Everything else is detail.