Cork MCP

Cork + AI - for Claude, the ChatGPT desktop app, GitHub Copilot, Gemini CLI, and any agent that speaks MCP

Unofficial. Community-built Claude Code Skill and MCP server for the Cork API. Not affiliated with, endorsed by, or sponsored by Cork. Cork and Cork Vantage are trademarks of Cork Protection, Inc.

[Cork - MCP server and Claude Code Skill]

_{Full skill page - install, outcomes, safety model.}

Every Cork API operation, plus cross-client risk attribution, exploitability-first vulnerability triage, overdue-compliance detection, and stale-connector health checks that a stateless API mirror cannot answer in a single call. Works with the AI you already use - ChatGPT (desktop app), Claude Desktop, Codex, Claude Code, Claude Cowork, and GitHub Copilot - plus Gemini CLI. Free, open source, runs on your laptop. Built for MSP owners. No code required.

Works with your agent

Security note. Local agents use the default stdio transport and need no network listener. HTTP mode (--transport http) refuses to start until CORK_MCP_HTTP_TOKEN is set, rejects any request without that bearer token, serves http://127.0.0.1:7777/mcp by default, and needs --tls-cert and --tls-key for any other address. It refuses a request body over 4 MiB or one that takes longer than 30 seconds to arrive. Whoever holds the token can call every tool with your Cork API key. Every tool that changes Cork data returns a preview and sends nothing until the call carries confirm: true. Set CORK_MCP_READ_ONLY=1 on any host that only needs to read: it removes every tool not marked read-only. See mcp-install.md.

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

The ChatGPT desktop app runs cork-mcp locally and shares MCP configuration with Codex (OpenAI). ChatGPT on the web needs a remote (HTTP) server. msp-skills does not support a web-only AI app yet (see mcp-install.md). If your team works there, a hosted gateway is the better pick; check that your AI app is on the gateway's setup list.

Also for the Microsoft and Google stacks

Big install base, but an honest heads-up: Gemini CLI runs the local binary you just installed, while Microsoft 365 Copilot and the Gemini app need a remote server, which msp-skills does not support yet.

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

Prefer the Claude Code plugin? Add the marketplace once, then install:

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

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

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

Install the Cork 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/cork/install.sh). If it uses Windows PowerShell, run iwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/cork/install.ps1 | iex. Then authenticate per the README and run cork-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/cork/install.ps1 | iex

macOS / Linux:

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

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

Verify:

cork-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/cork/install.sh)

Windows (PowerShell):

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

Claude Code plugin users: /plugin update cork@msp-skills.

Add to Claude Desktop, the ChatGPT desktop app, GitHub Copilot, Gemini CLI, 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 ChatGPT desktop app. 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/cork --force

Inside a Hermes chat session:

/skills install servosity/msp-skills/skills/cork --force

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

Install for OpenClaw

Tell your OpenClaw agent (copy this):

Install the cork skill from https://github.com/servosity/msp-skills/tree/main/skills/cork. The skill defines how its required CLI (cork-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 API key the CLI needs (Cork portal > Settings > API Keys):

export CORK_API_KEY=<your key>cork-cli doctor

doctor confirms the credentials work before you run anything that touches data. It exits 0 even when the credential is rejected, so scripts should add --fail-on error.

One property of Cork keys is worth knowing up front: an API key inherits the permissions of the user who created it. A 403 on the distributor or integration endpoints usually means the key was minted by an operator without that scope, not that the key is wrong. That property is also your best control - mint a read-only key for read-only work. See governance.md.

Sync before you ask

search, score regressions, and the client-roster lookups inside several other commands read a local mirror. Populate it once:

cork-cli sync

sync is read-only against the Cork API (GET requests only) and writes to a SQLite file under your own user account. Re-run it whenever you want fresher data; commands tell you when the mirror is stale rather than answering from data you did not expect.

What this skill does

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

Cork's own tooling is a stateless mirror of its REST API: it answers one client, right now. Every question an MSP actually asks is cross-client, cross-time, or both, and the Cork API exposes no cross-client aggregate endpoint at all, so the live path for any of them is a fan-out across every client with no stored yesterday to compare against.

This skill syncs Cork into a local SQLite mirror and does the fan-out and the joining for you, so "the risk score dropped" becomes "the risk score dropped because compliance moved, and here are the overdue events behind it" - one command instead of a per-client walk you assemble by hand.

Only score regressions answers purely from the local mirror. The other seven fetch from the Cork API on every run, because the rows they need carry no id that can be mirrored - vulnerability findings, compliance events, connector state, and score history. Several of them do use the mirror for the client roster so they can skip a live /clients fan-out, and fall back to a live roster scan when the mirror is empty. Every one of them caps its own scan and prints when a sweep was truncated, instead of reporting a false all-clear.

The pain this closes

• A score drops eleven points between QBRs. Cork returns the four components per timestamp but never differences them, so someone exports two snapshots and subtracts by hand. cork-cli score attribute does the subtraction and ranks the cause.
• "Which clients got worse this week" has no endpoint. cork-cli score regressions answers it as one local query.
• The vulnerability endpoint's server-side sort_by accepts only sw_vendor and sw_product, so exploitability ordering is impossible upstream. cork-cli vulnerabilities triage ranks known-exploited first, then EPSS, then CVSS, with a blast-radius device count.
• cve_id exists only nested inside cves[] and no endpoint accepts a CVE filter at any page size. cork-cli vulnerabilities exposure answers "are we exposed to this one" directly.
• A connector reports connection_status: ok while last_synced_at is days old. Both fields ship in the same payload and nothing compares the timestamp to now. cork-cli integrations health catches the green-but-dead connector.

See pain-point.md for the longer narrative.

Frequently asked questions

Does this work with ChatGPT?

Yes, in the ChatGPT desktop app. It runs MCP servers locally (STDIO), the way cork-mcp runs, and shares MCP configuration with Codex (OpenAI): add cork-mcp under Settings > MCP servers as a STDIO server, or add it once to Codex's ~/.codex/config.toml. ChatGPT on the web is not supported by msp-skills yet: it needs a remote (HTTP) server; see mcp-install.md. If your team works there, a hosted gateway is the better pick; check that your AI app is on the gateway's setup list.

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 Cork API key once.

Is my Cork data safe?

Your data stays on your machine. The CLI and MCP server are local binaries, and the MCP server speaks stdio by default - it opens no network listener unless you start its token-protected HTTP mode yourself. 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.

Will this hit my Cork API rate limits?

Partly. sync fills a local SQLite mirror, and search and score regressions answer from it without touching the API. export is NOT one of them - it paginates the live API and is one of the heaviest readers here, so pass --limit unless you mean to pull everything. The other seven analysis commands fetch live on every run, and two of them fan out - coverage gaps per connector, compliance overdue per client - so those are the ones to watch. Each caps its own scan (--max-clients, --max-connectors, --max-scan-pages) and tells you when it truncated. Cork publishes no rate limits, which is exactly why those caps exist.

Why does my Cork key get a 403 on some commands?

A Cork API key inherits the permissions of the user who created it. A 403 on the distributor or integration endpoints usually means the key was minted by an operator without that scope, not that the key is wrong.

Will this replace the Cork portal?

No, it complements it. The portal is still where you configure integrations and watch a single client. This skill answers the cross-client questions the portal makes you click through - score attribution, regression ranking, exploitability triage, stale connectors, coverage gaps - from your AI.

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

Three of those reach outside Cork itself: software install changes the state of a real customer machine, distributor provision-partner creates a partner account, and import runs either of those endpoints in bulk, one POST per line. Treat them like a manual RMM push or a commercial change, with the approval you already require for those.

The strongest control is the scope you grant the Cork API key - the CLI can only do what the key's creating user is permitted to do. Full details, including how to lock it down, are in governance.md.

Status

Beta. The full command surface has been exercised live, read-only, against a production Cork tenant: auth and connectivity, a complete sync, all eight cross-client analysis commands, typed exit codes on a missing resource, and the MCP stdio handshake. Mutating commands were run under --dry-run only; no write call was made against the tenant. Being validated further with MSPs running it against their own tenants 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: 2026-08-14.