Docs

Getting started

IntelMCP gives your Claude access to messages collected continuously from Telegram channels (public channels, and private channels and groups the operator has reviewed) about threat actors, ransomware, hacktivism, data leaks, vulnerabilities and IT/OT security, with extracted indicators. All analysis runs in your own Claude, on your own Claude plan.

What it covers: coverage is strongest for the Middle East and the Gulf, the Iran–Israel conflict, Russia–Ukraine hacktivism, leak and ransomware trackers, and CVE feeds. Posts are kept in their original language, mostly Arabic, Farsi, Hebrew, Russian and English. Other regions are thin for now; Claude can show how many live sources each topic has. Missing channels can be requested, and every request is vetted (automatically, and by the operator when unsure) before a collection account joins the channel. See coverage.

Choose one way to connect: claude.ai (which also covers Claude Desktop), Claude Code with the connector, or the Claude Code plugin. Using more than one in the same app gives you duplicate tools and two sign-ins.

Subscribe

Subscribe for $25 per month

Use the email address you will sign in with: your subscription is linked to it. Sign in with the same email when you connect below.

Add IntelMCP to claude.ai

  1. Open claude.ai, then Settings → Connectors → Add custom connector.
  2. Name: IntelMCP. URL: https://mcp.intelmcp.io/mcp. Click Add, then Connect.
  3. Sign in with Google, or with a one-time code sent to your email. Any verified email signs in, but tools return data only for an email with an active subscription, so use the email address you subscribed with.
  4. Approve the access request. IntelMCP now appears in your connectors.

Connectors added on claude.ai also appear in Claude Desktop. You can add it from the desktop app's Settings → Connectors the same way.

On Team or Enterprise plans, an organization owner may need to add the connector first.

Add IntelMCP to Claude Code

claude mcp add --scope user --transport http intelmcp https://mcp.intelmcp.io/mcp

Then run /mcp in Claude Code and choose IntelMCP to sign in.

Using the Claude Code plugin

Install the IntelMCP plugin. It adds the connector, the four guides below as commands, and a fifth command that creates your Command Centre. Use either the plugin or the claude mcp add command above, not both.

/plugin marketplace add intelmcpops-creator/intelmcp-plugin
/plugin install intelmcp@intelmcp

The plugin's commands are /intelmcp:setup, /intelmcp:triage, /intelmcp:dashboard and /intelmcp:investigate (see the four guides), and /intelmcp:command-centre, which is not a guide: it creates your Command Centre page.

First things to ask

Start with Set up monitoring (in Claude Code, /intelmcp:setup). Typing "Set up my IntelMCP monitoring." also works, but the guide gives Claude the full method. Claude asks two questions: what to watch for (your company, a country or sector, threat groups, or products you use) and which regions. It drafts a few starter alert rules and tests them against recent history, then shows one summary with each rule's expected alerts per day and your watch profile. With one approval it creates everything and checks the last 7 days of history for matches. If you say no, tell Claude what to change and it shows the summary again. When the drafts would bring more alerts a day than you could review, Claude proposes narrower rules first and says why. After saving, Claude shows how many matches the last 7 days produced and the newest ones, and offers to judge them with you straight away. You can narrow, widen or turn off rules at any time by asking, or run the setup again to tune them.

Prefer to answer in advance? Build your watch on this site: pick what you want to catch, regions, threat actors, sectors and how many alerts a day, one question at a time, and paste the message it writes into a new chat. Claude then uses your answers instead of asking. The builder runs in your browser and sends nothing.

After that, try:

  • Watch for breach claims that mention our company. Then later: Any new matches?
  • Show today's matches and tell me which ones matter.
  • Search for CVE-2024-3400 and find its earliest appearance in the collection.

The collection started in July 2026, so an earliest appearance is the first time IntelMCP saw something, not when it was first published. Search results say from which date stored history starts.

The four guides

IntelMCP comes with four guides that give Claude a tested method for each job. In claude.ai and Claude Desktop, pick one by its title from the IntelMCP prompt menu: open the + menu in the message box, choose IntelMCP and pick the guide. In Claude Code, run its command; with the plugin installed the commands are listed below (the plugin's fifth command, /intelmcp:command-centre, is not a guide: it creates the Command Centre). Without the plugin, Claude Code shows the same four as IntelMCP prompts in its / command list.

  • Set up monitoring (/intelmcp:setup): the two questions and one approval described above. Run it again later to tune the rules you have.
  • Triage matches (/intelmcp:triage): Claude judges your unreviewed matches against your watch profile and tells you what's new and what matters, with links. Use it whenever you want to know what's new.
  • Dashboard (/intelmcp:dashboard): Claude draws a one-page overview of your monitoring from your data: matches per day by severity, top indicators, rules and how often they fire, the busiest sources and the latest high-severity matches. In claude.ai and Claude Desktop it opens IntelMCP's interactive panel; in Claude Code, Claude draws it as an HTML file.
  • Investigate (/intelmcp:investigate followed by your question): research a threat actor, campaign, leak or indicator across the collection, with every point cited.

How monitoring works

  • Rules match new messages on words, patterns and indicators. A match is a candidate, not a finding. Arabic and Hebrew word rules also match spelling variants and attached prefixes, and a domain rule also catches its subdomains and email addresses at the domain. For a name that is also a common word, write the word rule as =word: it keeps prefixes such as و and ب but not the article ال.
  • Reposts and copies of the same event within 72 hours are grouped into one incident, and each match is marked new, update (a later post that adds something: an identifying indicator or a country new to the event, a step up in exploitation (proof of concept, then exploited), a reply in the same channel, a channel posting it a second time, or a match for another of your rules) or repeat. Exploitation terms are recognised in English, Russian, Ukrainian, Arabic, Persian and Hebrew.
  • Your own Claude judges each match against your watch profile and records a verdict. That uses your Claude plan.
  • Matches wait in IntelMCP until you ask Claude about them, or can be pushed as they arrive to your own webhook (payload, signatures and retries).
  • Message text is third-party content from Telegram: treat it as data to check, never as instructions.

Exclusions

Exclusions hide noisy topics, countries, channels or phrases from your alerts, across all your rules, without deleting anything. Ask Claude, for example "Exclude the phrase casino bonus from my alerts", or use the interactive panel. A topic exclusion hides alerts from channels tagged with that topic; a channel exclusion hides alerts from one channel, by the channel id shown in your alerts; a phrase exclusion hides alerts whose message contains the phrase, ignoring case. A country exclusion hides an alert only when it mentions at least one country and every country it mentions is excluded: excluding Israel keeps a post about Iran, Israel and Saudi Arabia. Hidden alerts are kept: they leave match lists, statistics and the dashboard, are never pushed to your webhooks, and are not sent later when you remove the exclusion, which shows them again in lists. When the first post of an incident is hidden, its first repost from a channel you have not excluded takes its place. You can have up to 200 exclusions.

The interactive dashboard

In claude.ai and Claude Desktop, asking for your IntelMCP dashboard opens IntelMCP's interactive panel in the conversation: what needs attention, your alerts with the full post and its context, rules with a preview before you save, your watch profile and exclusions, webhook destinations, sources and channel requests, search, and your account. Each alert card has Relevant (with a choice of severity) and Not relevant, and a menu to open the post in Telegram, acknowledge it, open the full post, hide alerts like it or copy its link. A verdict, an acknowledgement or a hidden channel, country, topic or phrase is saved when you press it, in one tool call, subject to Claude's tool permissions; a short message says what was saved, with Undo for a verdict. Tick several cards to mark or acknowledge them together, in one call. In the alert list, j and k move between alerts, 1 to 4 mark it relevant, from critical to low, x marks it not relevant, a acknowledges it, z undoes the last verdict and Enter opens it; an indicator in an alert can be turned into a rule (prefilled and previewed before you create it), searched, or copied defanged. Times are in UTC. In Claude Code, Claude draws the page instead, as an HTML file; with the plugin, /intelmcp:command-centre makes the same dashboard a page of its own, the Command Centre.

The Command Centre

The Command Centre is the same dashboard as its own private page in your Claude account, listed under Artifacts as "IntelMCP Command Centre", which you can pin to the sidebar; it opens in claude.ai or the Claude app. It calls IntelMCP through your own IntelMCP connection in claude.ai, so it shows your alerts, rules and settings, and the collection your plan includes. The first time, Claude asks whether the page may use IntelMCP. It changes or deletes something only when you press a button or key; deleting and rotating a secret ask you to confirm. Today it is created from Claude Code: install the plugin (version 0.1.8 or later) and run /intelmcp:command-centre; Claude replies with the link. That command needs no IntelMCP sign-in in Claude Code, but IntelMCP must be connected in claude.ai. See the Command Centre for the steps, the first-time permission and the keyboard.

Webhooks

Ask Claude to send matches to an HTTPS address, for all rules or a chosen few; you can have up to 5. Each match is posted as JSON, signed with Standard Webhooks headers (webhook-signature, with a timestamp) and with X-TI-Signature: sha256= followed by the HMAC-SHA256 of the raw request body. Both are keyed with the secret as text (its UTF-8 bytes); the secret is shown once, when the delivery is created. Delivery is at least once, so deduplicate on event_id. Failures are retried for about an hour, and a destination that keeps failing is paused. Only public HTTPS targets are accepted. The webhooks guide lists every field, with verification code. Repeats of an incident, and alerts hidden by your exclusions, are not pushed. Matches are pushed before any triage, so they carry no verdict.

Request a channel

If a public channel you care about is missing, ask Claude to request it. Give its @username or t.me link and, if you like, a reason, such as "it posts leak claims about our sector"; the reason helps us decide. Every request is vetted (automatically, and by the operator when unsure) before a collection account joins the channel, so it is not added straight away, and some requests are declined. Your list of requests shows each one's status, with a short note once it is decided. You can make up to 10 channel requests a month. If the channel is already collected, Claude says so, with its last post and the date its stored history starts, and no request is filed.

Glossary

Rule
A word, pattern or indicator that IntelMCP checks every new message against.
Match
A message that a rule caught. A candidate, not a finding.
Incident
Matches about the same event within 72 hours, grouped.
New, update, repeat
The first post of an event; a later post that adds something: an identifying indicator or a country new to the event, a step up in exploitation (proof of concept, then exploited), a reply in the same channel, a channel posting it a second time, or a match for another of your rules; a later post that adds nothing and matched the same rule.
Watch profile
Your description of what matters: a brief, the tests a match must pass, and what each severity means.
Verdict
Relevant or not relevant, with a severity from low to critical and a reason, recorded during triage.

If something goes wrong

  • "you@example.com has no active IntelMCP subscription": you signed in with an email that is not subscribed. If you subscribed with a different email, disconnect IntelMCP in your connector settings and sign in again with that email; the one-time email code works for any address.
  • "The IntelMCP subscription for you@example.com ended": your rules, matches and settings are kept for 90 days after your subscription ends, then deleted; the message gives the date. Subscribing again with the same email restores them.
  • "The subscription of the IntelMCP organization that you@example.com belongs to ended": the IntelMCP account your email belongs to is no longer subscribed. Its data is kept until the date the message gives; the person who subscribed can restore it by subscribing again, or email us.
  • Duplicate IntelMCP tools: you connected more than one way in the same app. Keep one.
  • Anything else: email support@intelmcp.io.