Integration

ChatGPT MCP for Bookmarks (2026 Setup Guide)

Beginner 5 minutes setup By David Hamilton

ChatGPT supports custom MCP connectors, which means you can point it at your ContextBolt library and search everything you have saved on X, Reddit, and LinkedIn from inside a conversation.

This is the setup where the difference shows up fastest, because chat is where you ask open questions. “What have I saved about this, and what are the common threads?” is a genuinely different question from a search box, and it is the one ChatGPT is good at.

Developer mode is the gate

Custom connectors sit behind Developer mode, and Developer mode sits behind a paid plan. Plus, Pro, Team, Enterprise, and Edu can all add one. Free ChatGPT cannot, and there is no workaround.

Open Settings, then Connectors (shown as Apps & Connectors on some accounts), then Advanced settings, and switch Developer mode on.

Adding the connector

In Settings, Connectors, click Create:

Authentication is None on purpose. Your ContextBolt token is already inside the URL path, so there is nothing for ChatGPT to authenticate separately. Picking OAuth or an API key here will stop the connector working.

The flip side is that the URL is your credential. Do not paste it into a shared chat, a screenshot, or a support ticket.

If it saves but finds nothing

MCP reads from your synced library. If cloud sync is switched off in the extension settings, the connector will validate, save, appear healthy, and return an empty result for every single query.

Turn sync on in the extension, give it a minute, and try again. It is worth checking this before you delete and rebuild the connector, because the symptoms point at the wrong thing.

What chat is actually good for

Synthesis across saves: “Summarize what I’ve saved about retention” reads across dozens of posts and gives you the themes. A search box cannot do that.

Grounded drafting: “Using my saved posts on cold email, draft a sequence” produces something shaped by material you already vetted rather than generic advice.

Finding the thing you half-remember: semantic search means you can describe the idea badly and still land on it.

Seeing what you actually collect: ask for your clusters and you get an honest picture of where your attention has been going.

This is separate from ChatGPT’s built-in memory, which only knows what you have told it in conversation. The two work together without conflict. If you want the fuller argument for why saved content beats chat memory, the ContextBolt bookmarks overview lays it out, and give Claude long-term memory makes the same case for a different client.

How to connect ContextBolt to ChatGPT

  1. Get your MCP token from ContextBolt

    Open the ContextBolt extension popup and click Settings. Under the MCP section, click Copy MCP Token. MCP is a Pro feature ($6/month) and cloud sync must be enabled in the same panel, otherwise the connector returns nothing.

  2. Turn on Developer Mode

    In ChatGPT open Settings, then Connectors (Apps & Connectors), then Advanced settings, and switch on Developer mode. Custom connectors need a paid plan: Plus, Pro, Team, Enterprise, or Edu. Free ChatGPT cannot add them at all.

  3. Add ContextBolt as a custom connector

    In Settings, Connectors, click Create. Name it Bookmarks, paste your MCP URL https://api.contextbolt.app/mcp/YOUR_TOKEN, and set authentication to None, because your token is already in the URL. Save.

  4. Verify it in a chat

    Start a new chat, confirm Bookmarks is toggled on in the tools menu, and ask 'search my bookmarks for anything about pricing strategy'. If it returns saves, you are live. If it returns nothing, check cloud sync in the extension before touching the connector.

Example prompts for ContextBolt + ChatGPT

Once connected, try asking ChatGPT:

Search my bookmarks for what I saved about pricing strategy

Semantic search across X, Reddit, and LinkedIn at once, so you describe the idea rather than recalling the exact post.

What have I saved recently about AI agents? Summarize the themes

Pulls recent saves and has ChatGPT synthesize across them, which is where a chat client beats a search box.

Using my saved posts on cold email, draft an outreach sequence

Grounds the draft in material you already curated and trust, rather than generic advice.

What are my bookmark clusters, and what's in the biggest one?

Chains list_clusters and get_cluster_bookmarks so you can see what you have actually been collecting.

What you can do with ContextBolt in ChatGPT

Search saved tweets, Reddit posts, and LinkedIn content by meaning

Summarize across many saves in one answer

Ground drafts in material you already curated

Browse your automatic topic clusters

Save new links to your library from a chat

Works alongside ChatGPT's other connectors

ContextBolt + ChatGPT: FAQs

Do I need a paid ChatGPT plan? +
Yes. Custom connectors require Developer mode, which is only available on Plus, Pro, Team, Enterprise, or Edu. Free ChatGPT has no route to adding a custom MCP connector at all. If you would rather not pay for ChatGPT, Claude Desktop and Claude Code both connect without a paid tier on the client side.
What authentication should I choose? +
None. Your ContextBolt token is already embedded in the MCP URL path, so there is nothing for ChatGPT to authenticate separately. Choosing OAuth or an API key here will break the connection. It also means the URL itself is the credential, so treat it like a password.
The connector saves but returns nothing. What's wrong? +
Almost always cloud sync. MCP reads from your synced library, so if sync is off in the extension settings the connector works perfectly and finds an empty library. Turn sync on, wait a minute for the upload, then retry. Check this before you rebuild the connector.
Can ChatGPT delete my bookmarks? +
No. It can search and retrieve, and it can create new bookmarks with save_bookmark. It cannot modify or delete anything already saved, so there is no risk to your library from a chat going off-script.
Is this different from ChatGPT's built-in memory? +
Completely. Built-in memory remembers things you have told ChatGPT in conversations. This connects it to content you deliberately saved from across the web, which is a much larger and better-curated body of material. The two work fine together.