Are you the author? Sign in to claim
Read X/Twitter threads, replies, quotes, and search through MCP using twscrape and your own session. No paid API.
An MCP server that reads X (Twitter): posts, threads, replies, quotes, and search. It wraps twscrape and uses your own logged-in session, so there's no paid X API and no developer account. Tools return clean markdown shaped for an agent to read, including image, video, and GIF URLs and any external links in a post.
Works with any MCP client over the two standard transports — local stdio and hosted Streamable HTTP.
| Tool | Returns |
|---|---|
read_tweet(url_or_id) | One post as markdown. |
read_thread(url_or_id, max_replies=50) | Root post + the author's self-thread + top replies. |
read_replies(url_or_id, limit=50) | Replies to a post. |
read_quotes(url_or_id, limit=30) | Quote-tweets (best-effort, search-based). |
user_timeline(username, limit=40, include_replies=False) | A user's recent posts, newest first. Set include_replies=True to include replies. |
search(query, limit=20, product="Latest") | Search results. Supports from:, has:media, min_faves:, etc. |
user_profile(username) | A user's profile as markdown: bio, location, follower/following/tweet counts, join date. |
uv tool install twscrape-twitter-mcp # or: pipx install twscrape-twitter-mcp
Then authenticate once (next section) and verify:
twscrape-twitter-mcp smoke # reads one public tweet end-to-end
Reads run against your own X session. Pick one path:
1. Launch a dedicated browser (recommended). Opens a separate Chrome/Brave profile with a DevTools port, you sign in to X once, and the session is captured. It does not touch your daily browser or automate X's login flow.
twscrape-twitter-mcp login --launch-browser chrome # or: brave
2. Attach to a dedicated browser profile. Only use this when you already run an isolated Chromium profile with a debug port. Browser debugging exposes browser data, and recent Chrome versions do not enable it for the default profile.
# macOS
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" --remote-debugging-port=9222 --user-data-dir=/tmp/x-mcp-browser
# Linux: google-chrome --remote-debugging-port=9222 --user-data-dir=/tmp/x-mcp-browser
# Windows: "C:\Program Files\Google\Chrome\Application\chrome.exe" --remote-debugging-port=9222 --user-data-dir=%TEMP%\x-mcp-browser
twscrape-twitter-mcp login --attach # add --cdp-url for a non-default port
3. Headless / CI (raw cookies). A server can't open your desktop browser, so
add a session from auth_token + ct0 cookies. Run this without cookie flags to
enter values through hidden prompts, rather than leaving secrets in shell history:
twscrape-twitter-mcp init
The captured session is reused across restarts. Run login --launch-browser chrome again when it expires,
or to add burner sessions for rate-limit rotation. twscrape-twitter-mcp accounts
lists the pool.
Use burner accounts, not your main — see Legal.
The server runs locally over stdio. Most MCP clients take a JSON block like this:
{
"mcpServers": {
"x": {
"command": "twscrape-twitter-mcp",
"args": ["serve", "--transport", "stdio"]
}
}
}
Client-specific equivalents:
claude mcp add x --scope user -- twscrape-twitter-mcp serve --transport stdio
Add the JSON block above to claude_desktop_config.json
(Settings → Developer → Edit Config).
~/.codex/config.toml[mcp_servers.x]
command = "twscrape-twitter-mcp"
args = ["serve", "--transport", "stdio"]
~/.cursor/mcp.json{
"mcpServers": {
"x": {
"command": "twscrape-twitter-mcp",
"args": ["serve", "--transport", "stdio"]
}
}
}
.vscode/mcp.json{
"servers": {
"x": {
"command": "twscrape-twitter-mcp",
"args": ["serve", "--transport", "stdio"]
}
}
}
For a hosted private instance (see Deploy), point a preconfigured client at the HTTP endpoint with a bearer token. This is static bearer auth, not an OAuth sign-in flow.
{
"mcpServers": {
"x": {
"url": "https://YOUR-APP.example.com/mcp",
"headers": { "Authorization": "Bearer YOUR_TOKEN" }
}
}
}
Then ask, e.g. "read this thread: <url>".
Run the server always-on and reachable over HTTP (POST /mcp). It ships as a
container; Cloudflare Workers won't work because twscrape is Python with native
deps.
A headless container can't open your desktop browser, so authenticate locally
first (login --attach writes storage_state.json under
TWSCRAPE_TWITTER_MCP_HOME), then ship that session to the host — copy the file to
the mounted volume, or run the cookie-based init over SSH. The server reloads a
persisted session on boot.
Always set a token when exposing HTTP — anyone who can reach the endpoint can use your X session:
export TWSCRAPE_TWITTER_MCP_AUTH_TOKEN=$(openssl rand -hex 32)
fly launch --no-deploy
fly volumes create twscrape_twitter_mcp_data --size 1
fly secrets set TWSCRAPE_TWITTER_MCP_AUTH_TOKEN=$(openssl rand -hex 32)
fly deploy
# seed a session onto the volume through hidden prompts:
fly ssh console -C "twscrape-twitter-mcp init"
Point Railway at this repo (it reads railway.json + Dockerfile), add a volume
mounted at /data, set TWSCRAPE_TWITTER_MCP_AUTH_TOKEN, and seed a session via
the Railway shell with twscrape-twitter-mcp init.
Clients send Authorization: Bearer <token>.
| Env var | Default | Purpose |
|---|---|---|
TWSCRAPE_TWITTER_MCP_HOME | ~/.config/twscrape-twitter-mcp | Where the sqlite account pool lives. Point at a volume in prod. |
TWSCRAPE_TWITTER_MCP_DB | $TWSCRAPE_TWITTER_MCP_HOME/accounts.db | Override the pool path directly. |
TWSCRAPE_TWITTER_MCP_AUTH_TOKEN | (unset) | Required bearer token for HTTP transport. |
TWSCRAPE_TWITTER_MCP_PROXY | (unset) | Global proxy for every account. |
TWSCRAPE_TWITTER_MCP_CDP_URL | http://127.0.0.1:9222 | Browser DevTools endpoint for login --attach. |
TWSCRAPE_TWITTER_MCP_DEFAULT_LIMIT | 40 | Default timeline and search result count (1–100). |
PORT | 8080 | HTTP port (Railway injects this). |
The hard part of reading X — GraphQL signing, the x-client-transaction-id
header, TLS fingerprinting — lives entirely in twscrape, which is pinned.
This package is a read-only MCP layer on top and never
touches that machinery. When X changes something and reads break, the fix is a
version bump, not reverse-engineering.
init.Reading X with your own logged-in session may violate X's Terms of Service, and accounts used for scraping can be rate-limited or suspended. Use burner accounts, not your main. Provided as-is for research and personal use; you are responsible for how you use it.
MIT.
The hard scraping work is twscrape by
vladkens. This is a read-only MCP layer on top — go star it.
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots
Google's universal MCP server supporting PostgreSQL, MySQL, MongoDB, Redis, and 10+ databases
Official GitHub integration for repos, issues, PRs, and CI/CD workflows