How to Search Your X Likes From Claude (MCP Guide 2026)
How to Search Your X Likes From Claude (MCP Guide 2026)
TL;DR: X Brain ships an MCP server, so Claude can search your liked X posts directly in a conversation. Generate a token in X Brain Settings, add the endpoint https://xbrain.live/api/mcp to your Claude MCP config with an Authorization: Bearer header, and Claude gains four read-only tools over your own library: semantic search, single-post lookup, category listing, and library stats. No other likes-archive tool currently exposes an MCP server.
Why connect your likes to Claude at all
Searching your archive in a web app is useful, but it puts you in the position of writing the query, reading the results, and doing something with them yourself. The point of connecting it to Claude is that your saved posts stop being a destination you visit and become context the assistant already has.
That changes the kind of question you can ask. Instead of "find posts about hiring", you can ask:
- "What have I saved about hiring engineers, and what's the common thread?"
- "I'm writing about onboarding. Pull anything relevant from my likes and quote the useful bits."
- "What topics did I save most in 2024 versus 2025?"
- "Find that thread I saved about database indexes and summarise it."
The last one is the case people hit most. You remember a post existed, you remember roughly what it argued, and you remember nothing searchable about it. Semantic search handles the recall; Claude handles the summarising.
What you need first
- An X Brain account with your archive already processed. The MCP server reads your processed library, so the upload and AI pipeline need to have run.
- Claude, in any client that supports MCP servers over HTTP.
Setup
Step 1: Generate an MCP token
Open Settings in X Brain and find the MCP integration section. Generate a token. It starts with xbrainmcp and is shown once, so copy it immediately. If you lose it, revoke it and generate a new one rather than trying to recover the old one.
Each token is tied to your account and can be revoked independently, so you can issue one per device and cut off any single one without disturbing the others.
Step 2: Add the server to Claude
Point Claude at the X Brain endpoint:
https://xbrain.live/api/mcpAuthentication is a bearer header:
Authorization: Bearer xbrain_mcp_your_token_hereThe server speaks JSON-RPC 2.0 over the Streamable HTTP transport, which is the standard remote-MCP setup, so any MCP client that supports remote servers with custom headers will work. The exact config file and syntax vary by client, and the Settings page shows the current endpoint alongside your token.
Step 3: Ask a question
Once connected, Claude lists the X Brain tools and calls them on its own when a question looks like it concerns your saved posts. You do not need to invoke tools by name. Asking "what have I saved about pricing strategy?" is enough.
The four tools
| Tool | What it does |
| --- | --- |
| search_tweets | Semantic search across your library. Matches by meaning, so "advice on firing someone" finds posts that never use those words. Optionally narrowed to one category or one year. |
| get_tweet | Fetches a single saved post by ID, including its extracted takeaway and references. |
| list_categories | Lists your categories with counts, which Claude uses to scope a follow-up search. |
| get_stats | Summary of the library: total posts, top categories, distribution by year. |
Everything is read-only. There is no tool that writes, edits, or deletes, so the worst a misfired call can do is return something you did not want to read.
What it cannot do
Two limits are worth stating plainly, because both come from the X archive format rather than from the integration.
You cannot filter by author. The like.js file in an X archive stores the post ID, its text, and a URL. There is no author field. Recovering authors would mean looking up every post ID through the X API, which is priced and rate-limited in a way that does not suit a one-time purchase. So Claude can find the post about database indexes, but it cannot answer "show me everything I saved from a specific account."
Dates are publication dates. The archive never records when you pressed like. What every post does carry is an ID that encodes when the post was published, so that is what the year filter uses. It is good enough for "narrow this to 2024" and wrong for "what did I save last spring."
For the full breakdown of what is and is not in the archive, see How to Use Your X Archive JSON File.
Security notes
- Tokens are scoped to a single account, and every query underneath is additionally protected by row-level security, so a token cannot reach another user's posts.
- The endpoint is rate-limited per token, and separately per IP before a token resolves, so a leaked or guessed token cannot be used to hammer the service.
- Revoke any token from Settings at any time. Revocation is immediate.
- Treat a token like a password. Anyone holding it can read your library.
Is this the same as ChatGPT plugins?
No. MCP is an open protocol for connecting assistants to external tools and data, and it is not tied to one vendor. X Brain implements the server half. Any MCP-capable client can connect, though the setup instructions here describe Claude because that is where most people will use it.
Frequently Asked Questions
Do I need a paid X Brain account to use MCP?
Yes. The MCP endpoint checks that the account is active before returning any data, the same as the rest of the AI pipeline. It is the same one-time $19 purchase, with no additional charge for the integration.
Does connecting to Claude cost extra in API credits?
Not from X Brain. Search runs against embeddings already stored in your library. You pay whatever your Claude usage normally costs, and X Brain does not bill per query.
Can Claude modify or delete my saved posts?
No. All four tools are read-only. There is no write path exposed over MCP at all.
Can other people see my library through MCP?
No. Every request resolves to your user account through your token, and the database enforces row-level security underneath that. A token only ever reaches its owner's posts.
What happens if I revoke a token while Claude is connected?
The next request fails with an authentication error. There is no grace period, which is the point: revocation is how you cut off a device you no longer trust.
Which other likes tools support MCP?
None that we are aware of at the time of writing. Most tools in this space are browser extensions or web apps with no programmatic interface, so this is currently specific to X Brain.