ThreatDown MCP

ThreatDown + AI - for ChatGPT, Claude, GitHub Copilot, Microsoft 365 Copilot, Gemini, and any agent that speaks MCP

Unofficial. Community-built Claude Code Skill and MCP server for the ThreatDown API. Not affiliated with, endorsed by, or sponsored by Malwarebytes Inc.

_{▶ Watch the 30-second demo - demo data is simulated; every command shown exists in the real CLI.}

Every ThreatDown OneView API, plus fleet-wide stale-endpoint, health and seat reports across all your sites, and guarded isolate/scan/restore jobs. Works with the AI you already use - ChatGPT (Plus/Pro+, paused until this connector is reprinted; see mcp-install.md), Claude Desktop, Codex, Claude Code, Claude Cowork, and GitHub Copilot - plus Microsoft 365 Copilot / Copilot Studio and Google Gemini via the remote path. Free, open source, runs on your laptop. Built for MSP owners. No code required.

Works with your agent

Security note (until this connector is reprinted). Do not run HTTP mode (--transport http), not even on localhost. This build checks a token, but it predates the fix release's request size and time limits. The MCP tools can also make changes with your credentials. Use the default stdio transport. If you set up a tunnel for a remote agent, shut it down. A fix is on the way.

The six agents MSP owners actually use (self-serve, works today):

Your AI agent	How to install the ThreatDown skill
Claude Desktop	Run installer, then Settings > Extensions to register threatdown-mcp (no JSON editing).
ChatGPT (paid plans)	Paused until this connector is reprinted: ChatGPT needs its HTTP mode, and in this build that mode checks a token but has no request size or time limits yet. See the security note in mcp-install.md.
Codex CLI	Paste the install prompt below.
Claude Code	Paste the install prompt below.
Claude Cowork	Paste the install prompt below.
GitHub Copilot (VS Code)	Run installer, add threatdown-mcp to mcp.json under the servers key, then pick Agent mode.

ChatGPT needs a remote HTTPS server, which means this connector's HTTP mode, and that is paused until the connector is reprinted: in this build that mode checks a token but has no request size or time limits yet. See the security note in mcp-install.md.

Also for the Microsoft and Google stacks

Big install base, but an honest heads-up: these are the remote / enterprise path, not the local binary you just installed.

Agent	What it takes
Microsoft 365 Copilot / Copilot Studio	Paused until this connector is reprinted: Copilot Studio needs its HTTP mode, and in this build that mode checks a token but has no request size or time limits yet. See the security note in mcp-install.md.
Google Gemini	Gemini CLI is local - same as Claude Code. The Gemini app is remote and needs the same paused HTTP mode as ChatGPT. See mcp-install.md.

Skill-native agents (also covered): Hermes and OpenClaw read this skill's SKILL.md directly and speak MCP - see their install sections below. Also works with Cursor, Windsurf, Cline, Continue.dev, and Zed via MCP. Full per-tool wire-up: docs/which-agent.md.

Run more than one agent? Install across all 51+ supported agents in one command: npx skills add Servosity/msp-skills@latest (requires Node.js, then run the per-skill installer for the CLI/MCP binaries). See docs/which-agent.md.

Install in 60 seconds

Fastest for Claude Desktop - one-click .mcpb

Download ThreatDown MCP (.mcpb) - then open Claude Desktop > Settings > Extensions and select the file. One click, no JSON, no shell. (Browse every ThreatDown release on the releases page.)

Prefer the Claude Code plugin? Add the marketplace once, then install - works immediately, no directory listing required:

/plugin marketplace add Servosity/msp-skills
/plugin install threatdown@msp-skills

Path A - paste one prompt into your AI agent (recommended)

Copy this into Claude Code, Codex CLI, or Claude Cowork:

Install the ThreatDown Skill and MCP server from Servosity/msp-skills in this agent workspace. If this workspace uses a POSIX shell (macOS, Linux, WSL, or Bash), run bash <(curl -fsSL https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.sh). If it uses Windows PowerShell, run iwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.ps1 | iex. Then authenticate per the README and run threatdown-cli --help to explore.

The same prompt works in any agent that can run shell.

Path B - run the installer yourself

Windows (PowerShell):

iwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.ps1 | iex

macOS / Linux:

bash <(curl -fsSL https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.sh)

The installer drops both threatdown-cli (the CLI) and threatdown-mcp (the MCP server) into your user bin path. Claude Code, Codex, and Cowork discover the Skill via SKILL.md in this directory.

Verify:

threatdown-cli --version

Upgrade to the latest version

The installer always fetches the current release - re-run it to upgrade:

macOS / Linux:

bash <(curl -fsSL https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.sh)

Windows (PowerShell):

iwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/threatdown/install.ps1 | iex

Claude Desktop .mcpb users: download the latest .mcpb (top of this section) and re-select it in Settings > Extensions. Claude Code plugin users: /plugin update threatdown@msp-skills.

Add to Claude Desktop, GitHub Copilot, Gemini CLI, Microsoft 365 Copilot, or another MCP client

After the installer runs, see mcp-install.md and docs/which-agent.md for the per-agent wire-up - one section per agent, including the GitHub Copilot servers key and the remote Microsoft 365 Copilot / Copilot Studio path. Claude Desktop's Settings > Extensions panel is the simplest path; the MCP config block (for users who prefer editing JSON) is documented in mcp-install.md.

Install for Hermes

From the Hermes CLI:

hermes skills install servosity/msp-skills/skills/threatdown --force

Inside a Hermes chat session:

/skills install servosity/msp-skills/skills/threatdown --force

Hermes speaks MCP natively, so it can also use the threatdown-mcp server directly - same install path, same env vars.

Install for OpenClaw

Tell your OpenClaw agent (copy this):

Install the threatdown skill from https://github.com/servosity/msp-skills/tree/main/skills/threatdown. The skill defines how its required CLI (threatdown-cli) can be installed via the openclaw: frontmatter block.

OpenClaw isn't generally available yet; the frontmatter wiring is pre-shipped and will activate the moment OpenClaw launches.

Authenticate

Set the credentials the CLI needs (from your ThreatDown portal):

export THREATDOWN_CLIENT_ID=<value>
export THREATDOWN_CLIENT_SECRET=<value>
threatdown-cli doctor

Keep both variables set for every process that runs the CLI or the MCP server (the MCP install puts them in your client's env block). doctor exchanges them for a token and probes it with an authenticated call to OneView's account endpoint (its Credentials row), so a wrong secret shows there as a credentials error with HTTP 401. Only the minted token is cached, in the CLI's credentials.toml, and a process without the variables does not reuse it. To store the pair instead, export both variables once and run threatdown-cli auth login --client-id <id>: it saves that client ID and the secret from the environment in that file until threatdown-cli auth logout, and the secret never appears on a command line.

What this skill does

Question your MSP keeps asking	Command
Which endpoints haven't checked in for 24 hours, across every site?	threatdown-cli fleet stale --hours 24 --agent
Which customer site needs attention first?	threatdown-cli fleet health --agent
Which sites are over- or under-allocated on seats, and which subscriptions are expiring?	threatdown-cli seats reconcile --agent
Which scans or isolations never ran?	threatdown-cli respond pending --since 7d --agent
Which threats hit more than one customer this month?	threatdown-cli threats spread --since 30d --min-sites 2 --agent
Which infected or suspicious machine should I contain next?	threatdown-cli threats queue --since 7d --agent
Everything about one customer while they're on the phone	threatdown-cli customer report "Acme Corp" --agent
Isolate a machine, with a preview first	threatdown-cli respond isolate --host LAPTOP-042 --agent

threats spread, threats queue and customer report read the local copy, so run threatdown-cli fleet sync first (and again when you want fresh numbers). The respond commands send nothing until you add --confirm.

Full command reference: guide.md. For the AI-agent operating contract (--agent, --dry-run, when to confirm before mutating), see AGENTS.md.

What makes this different

We found no other MCP server for ThreatDown OneView on GitHub as of October 2026. The integrations that do exist, such as the Sumo Logic and Datadog (Crest Data) apps, move OneView data into a SIEM or monitoring tool. A question like "which threats hit more than one of my 40 customers this month" is still a per-site export and a spreadsheet.

This skill resolves every site's account ID for you and syncs sites, endpoints, detections and jobs into a local SQLite mirror with full-text search (search --data-source local). Aggregate questions become one local SQL join: instant, offline, and the AI sees the answer, not the raw data. Compound commands like threats spread, threats queue and customer report join detections, endpoints, jobs and sites - work a stateless API wrapper can't do. ThreatDown's own AI features, such as the AI-assisted Security Advisor and AI Detection & Response, work inside the ThreatDown console; this skill complements them by bringing every OneView site into the agent you already use.

The pain this closes

ThreatDown OneView answers one site at a time. The Monday question - "which machines went dark since Friday, across all my customers?" - has no single page. ThreatDown's own answer is a script: the support article "API Example: Get endpoints not seen for more than 24 hours" (article 35082574965011) ships PowerShell that lists every site, collects each account ID, pages the endpoint search and prints hours since last seen. The dashboard's "Last sync 7+ days ago" widget catches a machine only after a week of silence. Site health, seat counts and job outcomes have the same per-site shape.

• threatdown-cli fleet stale --hours 24 - the vendor's PowerShell example as one command across every site.
• threatdown-cli fleet health - one ranked row per customer site, worst first.
• threatdown-cli seats reconcile - allocations against deployed and active machines, with expiring subscriptions flagged.
• threatdown-cli respond pending --since 7d - the scans, isolations and RMM-issued jobs that never ran.
• threatdown-cli respond isolate --host LAPTOP-042 - finds the machine across every site and previews before sending.

See pain-point.md for the longer narrative.

Frequently asked questions

Does this work with ChatGPT?

Not with this build. ChatGPT (Plus, Pro, Team, Business, Enterprise, and Education plans; Free does not yet expose Developer Mode) connects only to remote MCP servers over HTTPS, which means running the ThreatDown MCP server in HTTP mode. In this build HTTP mode checks a token but has no request size or time limits yet, so do not run it or put it behind a tunnel until the connector is reprinted; the security note in mcp-install.md explains why. Until then, use Claude Desktop, Claude Code or Codex.

Does this work with Codex, Cursor, Windsurf, Cline, Copilot, or Gemini?

Yes - all of them speak MCP. Cross-tool install commands are in the matrix above and the deep-dive in docs/which-agent.md.

Do I need to know how to code?

No. The recommended install is to paste one sentence into Claude Code or Codex - your agent reads SKILL.md and does the install. The fallback is a one-line installer per OS (bash or PowerShell). Neither path requires writing code. You'll enter your ThreatDown credentials once.

Is my ThreatDown data safe?

Your data stays on your machine. The CLI and MCP server are local binaries. The SQLite mirror sits in a directory under your user account. The AI agent only sees what the CLI returns - typically a query result, not raw bulk data. Credentials are read from your environment or your agent's config; never bundled into this repo or transmitted anywhere by MSP Skills.

Is this for ThreatDown OneView or ThreatDown Nebula?

OneView, the multi-site console ThreatDown (Malwarebytes) gives MSPs. It talks to the OneView API at api.threatdown.com/oneview and works across every site you manage. A single-company Nebula console is not a OneView account; it uses a different API, which this skill does not cover.

What credentials do I need?

An OAuth client ID and secret from OneView: open Integrate in the left menu, click Add, name the app and tick the scopes. ThreatDown says a Global or Site Administrator has to create it, and the secret is shown only once. read covers every report; the respond commands (isolate, restore, scan, reboot, update) also need execute, and other changes need write. Set THREATDOWN_CLIENT_ID and THREATDOWN_CLIENT_SECRET, then run threatdown-cli doctor. To request fewer scopes than the default execute read write, set THREATDOWN_OAUTH_SCOPE (for example read); the CLI requests a token with those scopes on its next call.

Can my AI isolate or scan client machines?

Only through commands you allow. The respond commands preview by default: they list the machines that would get the job and send nothing until you add --confirm (or the global --yes), and they refuse more than 25 endpoints unless you raise --max-targets. Other paths send immediately: endpoint-jobs issue, endpoint-jobs bulk-issue, grid jobs-bulk, import endpoint_jobs, and the MCP server's generic threatdown_execute tool. For a reporting agent, create the OAuth client with read only, or set THREATDOWN_MCP_READ_ONLY=1 so the MCP server drops its write tools and refuses every non-GET call, and endpoints diag-log, through threatdown_execute.

Has it been tested against a live OneView tenant?

Not yet. Every command was built from ThreatDown's published OneView API specification and tested against a local mock of that API, not a real tenant. If you run it against your own OneView account, tell us what worked or broke on the issue tracker or at a Build Session; a real MSP run is what earns the live-verified badge.

Will this replace the OneView console?

No. OneView stays where you set policies, manage subscriptions and deploy agents. This skill answers cross-site reporting and triage questions from your AI agent or terminal, and sends response jobs: the respond commands preview first, while the raw job commands send immediately.

What does it cost?

Free. Apache-2.0 licensed. You pay only for whichever AI agent you use (Claude, ChatGPT, Codex, etc.), and that's billed by your AI provider, not by us.

Safety model

Tier	Examples	Recommended agent policy
Read	fleet stale, fleet health, seats reconcile, respond pending, threats spread, threats queue, customer report, fleet sync, search, tail (it strips the secret fields), the respond commands without --confirm, every list / get / search / export command except the credential reads in the last row and endpoints diag-log	Allow
Endpoint actions	respond isolate / restore / scan / reboot / update with --confirm, endpoint-jobs issue, endpoint-jobs bulk-issue, endpoint-jobs quarantine-bulk, endpoint-jobs parent-reissue, endpoint-jobs parent-children-reissue, the endpoint-jobs cancels, grid jobs-bulk, endpoints sa-remediate, suspicious-activity remediate, endpoints diag-log, import endpoint_jobs, the MCP threatdown_execute tool on a jobs endpoint	Human-in-the-loop, explicit confirmation
Security config	exclusions create / update, policies create / update / promote and the other policy writes, groups assign-endpoints, firewall, content-filtering, device-control, app-block and AIDR rules, schedules, webhooks and notifications create / update, syslog and email-protection settings, and every other write not in the rows around it	Preview with --dry-run, then a human-approved write
Destructive	endpoints delete / delete-bulk, grid endpoints-delete, sites delete, policies delete, exclusions delete, users delete, webhooks delete, notifications delete, and every other delete command	Human-in-the-loop only, explicit confirmation
Credential and admin	Every policies read (including secret-get / secret-get-v2), sites secret, installation-tokens search, installers, webhooks list / get, notifications list / get, and export of policies or notifications (their output can carry tamper-protection passwords, enrollment credentials or webhook signing keys; the local mirror never stores those fields), installation-tokens create / send / revoke, policies secret-set, users create / update, sites create / update, import of users or sites, subscriptions create / update, the email-protection register, migrate and authorize commands	Human-in-the-loop only

The strongest control is the scope you tick on the OneView OAuth client (read, write, execute) - the CLI can only do what that client is permitted to do. Set THREATDOWN_MCP_READ_ONLY=1 to run the MCP server read-only. Full details, including how to lock it down, are in governance.md.

Status

Beta. Built from ThreatDown's published OneView API specification and tested against a local mock of that API; being validated with MSPs running it live against their own production tenant in our weekly Build Sessions. RSVP at compoundingteams.com/build-sessions.

---

Standards. Conforms to the open Agent Skills spec (Anthropic, Dec 2025; 40+ agents). MCP-compatible - works with any MCP-capable agent including Hermes. OpenClaw-ready (frontmatter pre-wired, awaiting OpenClaw launch).

Maintained by Servosity. Apache-2.0 licensed. Built with CLI Printing Press. Last updated: 2026-10-14.