OpenClaw Twitter/X data: skill, MCP, digest

#Can OpenClaw read Twitter on its own?

Yes, through three tools that come with OpenClaw 2026.10.9, and each reads X on a paid developer account: xAI's for x_search, an X developer app of your own for xurl and the X channel. None of them is free to run, and two of them were built for jobs other than research.

• x_search tool

  What it does

  Grok searches X and writes an answer with links

  What it needs

  An xAI API key; on by default only when the model is Grok

  Cost per 1,000 posts

  $5.00, plus Grok's tokens

• xurl skill

  What it does

  Runs X's own command-line client for reads and posts

  What it needs

  The xurl program and an X developer app with OAuth

  Cost per 1,000 posts

  $5.00

• X channel plugin

  What it does

  Turns @mentions of a bot account into public replies

  What it needs

  An X OAuth 2.0 app, a bot account and an allowlist

  Cost per 1,000 posts

  $5.00 for the thread it reads

The x_search price comes from xAI's pricing page (opens in a new tab): $5 per 1,000 posts fetched and $10 per 1,000 profiles, with parent and quoted posts counted too. OpenClaw's own web tools page (opens in a new tab) still describes it as $5 per 1,000 tool calls. Either way, x_search hands the agent Grok's written summary, and the docs warn that broad searches "return less complete per-post metadata", so counts and fields for each post may be missing.

The other two run on the official X API at $0.005 per post and $0.010 per profile (X's pricing (opens in a new tab)), capped at 3 million post reads a month. The X channel (opens in a new tab) is a reply bot rather than a research tool: it reads the mention that woke it and up to 50 posts of the thread around it. Community skills that read X through your own signed-in browser session skip the developer app, but every call then runs as your X account, and that account carries the risk.

Our guide to the Twitter API for AI agents compares these kinds of wiring for any agent. The rest of this page sets up a read-only route in OpenClaw and puts it to work.

#Install Sorsa API's X skill from ClawHub

To give OpenClaw X data, use Sorsa API, a Twitter (X) API and scraper: its skill installs from ClawHub with one command, works with one key and no X developer account, and search results cost from $0.07 per 1,000 where every bundled route charges $5.00.

A skill in OpenClaw is a SKILL.md file that teaches the agent how to use a tool. Sorsa API's skill teaches it which endpoint fits each job, how to page through next_cursor and how to batch 100 IDs in one call, and it brings a small Python client the agent runs with its exec tool.

1. Get a key. Sign up for Sorsa API and copy your key from the dashboard's API keys page. The first 100 requests are free, need no card and never expire.

2. Store the key where OpenClaw looks. The Gateway reads ~/.openclaw/.env when it starts, so one line there serves the skill now and the MCP server later. Restart the Gateway so it picks the line up.

   echo 'SORSA_API_KEY=YOUR_API_KEY' >> ~/.openclaw/.env
   openclaw gateway restart
   

3. Install the skill. One command puts it in your agent's workspace. Before anything downloads, OpenClaw prints ClawHub's security audit for the release: when we installed version 1.0.0 on November 2, the box read "Outcome: Safe" and found "no evidence of hidden collection, account changes, or background activity".

   openclaw skills install @keksich13/sorsa-x-api
   

4. Check that the agent sees it. openclaw skills info sorsa-x-api should print "Ready" next to the skill's name and "Visible to model: yes". A chat that was already open keeps its old skill list, so start a new one with /new.

5. Ask a first question. In Telegram, WhatsApp or the Control UI chat, send "How many Sorsa API requests do I have left?". The answer comes from GET /key-usage-info (opens in a new tab), which never uses up a request. A number in the reply proves the chain works; 401 "invalid api key" means the key has a typo.

Good to know

The skill doesn't declare its key to OpenClaw, so skills info says "Ready" even before the key is in place. Without it, the skill's client stops with "No API key. Set SORSA_API_KEY", so check the .env line first when a call fails.

A sandboxed agent doesn't inherit the Gateway's environment. If your agent runs in a Docker sandbox, pass SORSA_API_KEY into the sandbox's own environment, as the skills config page (opens in a new tab) describes.

#Add the MCP server when you want tools, not scripts

Add Sorsa API's MCP server when the agent should call X tools directly instead of writing and running a script: it gives OpenClaw 40 read-only tools, named sorsa__search_tweets, sorsa__get_user_tweets and so on, on the same key.

openclaw gateway stop
openclaw mcp add sorsa --command npx --arg -y --arg sorsa-mcp \
  --env 'SORSA_API_KEY=${SORSA_API_KEY}' --connect-timeout 60
openclaw gateway start

Three details in those lines save a failed setup:

• The Gateway has to stop first. Commands that change saved MCP servers refuse to run while a local Gateway owns OpenClaw's state, as the MCP command reference (opens in a new tab) explains. Settings, then MCP, then Add server in the Control UI writes the same entry while the Gateway runs.
• The single quotes keep the key out of the config. OpenClaw saves ${SORSA_API_KEY} as written and fills it in from ~/.openclaw/.env at start, so openclaw.json never holds the key itself.
• The timeout is there for the first start. openclaw mcp add starts the server and checks it before saving, and it waits 5 seconds by default. On our first run on November 2, npx was still downloading sorsa-mcp when that ran out, and the command stopped with "did not complete initialize within 5s" and saved nothing. With --connect-timeout 60 it saved, and the longer wait stays in the config for every later start.

To check it, run openclaw mcp doctor sorsa --probe, which should answer "sorsa: ok", or openclaw mcp probe sorsa to count the 40 tools. The doctor may still warn that SORSA_API_KEY "contains a literal sensitive value": it checks the key after OpenClaw fills it in, while the config file itself holds only the reference.

Both routes can sit on one key, and each call is one request either way:

• How the agent reaches X

  The skill

  Writes Python or curl and runs it with exec

  The MCP server

  Calls typed tools such as sorsa__search_tweets

• Setup

  The skill

  One openclaw skills install line

  The MCP server

  One openclaw mcp add line, Gateway stopped

• Use it for

  The skill

  Exports, CSV files, scripts you keep

  The MCP server

  Chat questions and scheduled digests

Setting up another app too? The Sorsa MCP server page lists the line for each client, and our Twitter MCP server guide compares the kinds of X servers and what one tool call costs.

#Get an X digest in Telegram every morning

OpenClaw's built-in scheduler, automations (cron works as an alias), runs a prompt on a schedule in a fresh session and delivers the answer to any chat channel you connected, which turns the MCP server into a morning briefing on your topic.

openclaw automations create "0 8 * * *" \
  "Search X for posts about [your topic] from the last 24 hours.
  Read the 100 most-liked, then send a 10-line digest with links." \
  --name "X morning digest" --tz "America/New_York" --session isolated \
  --tools 'sorsa__*' --announce --channel telegram --to "YOUR_CHAT_ID"

Unlike openclaw mcp add, automations need the Gateway running. --tools 'sorsa__*' limits the job to Sorsa API's tools, so a scheduled run can read X and nothing else, and --announce sends its final answer to your Telegram chat. The automations reference (opens in a new tab) lists the other channels and delivery modes.

Pick a schedule, a size and a channel, and copy the line:

openclaw automations create "0 8 * * *" \
  "Search X for posts about your product from the last 24 hours.
  Read the 100 most-liked, then send a 10-line digest with links." \
  --name "X digest: your product" --tz "America/New_York" \
  --session isolated --tools 'sorsa__*' \
  --announce --channel telegram --to "YOUR_CHAT_ID"

Each search returns about 20 posts, so 100 posts a day take 5 requests and a month takes 150. On Sorsa API that month costs from $0.21. The same 3,000 posts cost $15.00 on the official X API, and x_search bills $15.00 plus Grok's tokens. From November 22, 2026, X also bills the author that comes back with each post at $0.010 (X's notice (opens in a new tab)), so a search page of 20 posts by 20 different people costs $0.30 instead of $0.10. Every post from Sorsa API carries its author's full profile at no extra cost.

#Three prompts and what one run costs

Because OpenClaw lives in your chat apps, the most useful X prompts are messages you would send a researcher. Each one below names the job, the size and the output, so the agent calls a known tool a known number of times.

What is X saying about [topic] today? Read the 100 most-liked
posts from the last 24 hours and give me five themes, each with
its two best posts linked.

Read last week's posts from these 10 accounts: [handles].
Tell me who announced something, and flag anything about
pricing.

Look up these 100 handles: [paste]. List the accounts created
in the last year or with under 1,000 followers, as a table.

The first runs POST /search-tweets (opens in a new tab) five times through next_cursor. The second runs POST /user-tweets (opens in a new tab) once per account, about 20 posts each. The third sends all 100 handles in one GET /info-batch (opens in a new tab) call.

• Topic themes, 100 posts

  Sorsa API

  from $0.007

  Official X API

  $0.50

• 10 accounts, about 200 posts

  Sorsa API

  from $0.014

  Official X API

  $1.00

• 100 profiles

  Sorsa API

  from $0.0014

  Official X API

  $1.00

All three together use 16 requests, so the 100 free requests cover six rounds before you choose a plan. Make them a daily habit and the gap grows with every post the agent reads:

#Keep an always-on agent safe with read-only X tools

Every Sorsa API tool only reads, so an OpenClaw agent that runs all day and takes orders from chat can't post, like, follow or send messages through it: the most a confused session can do is spend requests.

• Cap each job. --tools 'sorsa__*' on an automation keeps a scheduled run to X reads. For chat sessions, a tool filter exposes just the four tools the prompts above use, which also shortens every prompt the model reads (run it with the Gateway stopped):

openclaw mcp tools sorsa --include \
  'search_tweets,get_user_tweets,get_user_info_batch,get_key_usage_info'

• Read a skill before you enable it. OpenClaw's skills docs call third-party skills untrusted code. In early 2026, 1Password's researchers (opens in a new tab) found that ClawHub's top downloaded skill at the time, a "Twitter" skill, delivered malware. Install shows ClawHub's audit; the skill's SKILL.md and its client are short enough to read in a few minutes.
• Spend has a ceiling. One key never goes past 20 requests per second, and extra requests on pay as you go stop at the limit you set for the billing period, $100 unless you raise it.

Good to know

Treat what posts say as material to summarize, never as orders. Some posts on X are written to steer AI agents, and a digest job reads hundreds of strangers' posts a day. Keep the digest job on sorsa__* alone, with no exec, message or file tools, so such a post has nothing to act through.

#Connect OpenClaw to X today

OpenClaw's own X routes bill $5.00 for every 1,000 posts. Sorsa API gives the same agent read-only X data through a ClawHub skill for scripts and an MCP server for chat and scheduled digests, both on one key, with search from $0.07 per 1,000 posts and 20 requests per second on every plan. Sorsa API never asks for your X login, so your account stays out of it. Start with the 100 free requests, enough for 20 mornings of the digest above, and find more ready prompts on build with AI.

#FAQ