接入你的智能体
發布者自己的指令。不是 PlugMCP 端點。
用我們的金鑰與日誌託管一份副本。
mcpb https://github.com/Servosity/msp-skills/releases/download/sherweb-v0.1.12/sherweb-mcp.mcpbClaude Code 沒有本機指令的安裝連結。複製下方指令並在終端機中執行。
SHERWEB_BASE_URL=<value> SHERWEB_CLIENT_ID=<secret> SHERWEB_CLIENT_SECRET=<secret> SHERWEB_OAUTH_SCOPE=<secret> SHERWEB_SUBSCRIPTION_KEY=<secret> SHERWEB_TOKEN_URL=<value> SHERWEB_NO_CONFIG_WRITE=<value> SHERWEB_MCP_READ_ONLY=<value> claude mcp add io.github.servosity-sherweb-mcp -- mcpb https://github.com/Servosity/msp-skills/releases/download/sherweb-v0.1.12/sherweb-mcp.mcpb在終端機中執行。
{ "mcpServers": { "io.github.servosity-sherweb-mcp": { "command": "mcpb", "args": [ "https://github.com/Servosity/msp-skills/releases/download/sherweb-v0.1.12/sherweb-mcp.mcpb" ], "env": { "SHERWEB_BASE_URL": "<value>", "SHERWEB_CLIENT_ID": "<secret>", "SHERWEB_CLIENT_SECRET": "<secret>", "SHERWEB_OAUTH_SCOPE": "<secret>", "SHERWEB_SUBSCRIPTION_KEY": "<secret>", "SHERWEB_TOKEN_URL": "<value>", "SHERWEB_NO_CONFIG_WRITE": "<value>", "SHERWEB_MCP_READ_ONLY": "<value>" } } }}claude_desktop_config.json
遠端 URL 還是本機指令？ 兩種安裝方式的差異
來自發布者
MSP Skills connects your PSA, RMM, backup, and more to the AI you already use - Claude, ChatGPT, Codex, Cursor, Windsurf, or any agent that speaks MCP. Ask a plain-English question about your stack and get a real answer back.
Every Sherweb Partner API capability, plus a local SQLite store, offline analytics, and
84 connectors are live today - including Servosity, ConnectWise PSA, HubSpot, and HaloPSA - and more PSA, RMM, backup, and M365 connectors ship every week.
Free, open source, runs on your laptop. A local SQLite mirror lets your agent answer cross-client questions the live API can't return in one shot - no rate-limit hits, no per-tech SaaS fee, and your data stays local unless you route it somewhere yourself. Built for MSP owners. No developer experience required.
New to the term? What this repo calls an MCP server is what ChatGPT calls an app or connector, Claude on the web calls a connector, Microsoft Copilot calls a connector, and Claude Code calls a Skill. Same standard underneath: the Model Context Protocol. Full plain-language answer: What is an MCP server?.
You do not need to know which command does what. Open Claude Code and paste:
Set up msp-skills for me from https://github.com/servosity/msp-skills. Read the repo's README, install the connectors that fit my stack, and walk me through anything only I can do (like approving a browser extension or signing in to a vendor portal).
Two notes that save the most time, on Windows and macOS alike:
/plugin marketplace add https://github.com/servosity/msp-skills.git
/plugin install connect-tool@msp-skills
.git suffix matters.irm https://raw.githubusercontent.com/servosity/msp-skills/main/skills/connect-tool/bootstrap.ps1 | iex
Stuck on credentials? That is what connect-tool is for: it drives the Chrome you are already logged into, puts the API key straight into Windows Credential Manager or the macOS Keychain, and refuses to say "connected" until a real authenticated call returns your live data.
⭐ First-party, by Servosity: the servosity backup & DR connector + the guided concierge and the connect-tool.
About the badges.
Live-verified(green) means a real MSP confirmed this skill against a live production tenant - the badge carries the date and source.Awaiting live verification(amber) means it passes every mechanical gate (build, command-surface-vs-docs claims check, install dry-run) but no one has reported a live run yet - be the first: run it against your tenant and report that it works.
🛠 Built live with MSPs. Join a free weekly Build Session at compoundingteams.com/build-sessions to watch a Skill built against a real MSP system - or bring your own.
Not sure which connector to pick? Let your agent decide. Paste this into Claude Code, Codex CLI, or Claude Cowork:
Read https://github.com/Servosity/msp-skills and, using everything you know about me and how my MSP works, recommend which connectors I should install - then install the ones I approve.
Prefer a guided version? The concierge does the same from inside Claude Code: /plugin marketplace add Servosity/msp-skills, then /plugin install msp-skills-concierge@msp-skills.
Or install a specific connector. The Servosity prompt:
Install the Servosity 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/servosity/install.sh). If it uses Windows PowerShell, runiwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/servosity/install.ps1 | iex. Then authenticate withSERVOSITY_MSP_TOKEN=<your-partner-token> servosity-cli doctorand runservosity-cli --helpto explore.
And the ConnectWise PSA (Manage) prompt:
Install the ConnectWise PSA (Manage) 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/connectwise-manage/install.sh). If it uses Windows PowerShell, runiwr -useb https://raw.githubusercontent.com/Servosity/msp-skills/main/skills/connectwise-manage/install.ps1 | iex. Then authenticate per the README and runconnectwise-manage-cli --helpto explore.
The same prompt works for every connector in the table above - swap the skill name and slug. If your agent can't run shell, use Path B below.
On macOS / Linux (swap the slug for any connector in the table above):
bash <(curl -fsSL https://raw.githubusercontent.com/servosity/msp-skills/main/skills/servosity/install.sh)
On Windows (PowerShell):
iwr -useb https://raw.githubusercontent.com/servosity/msp-skills/main/skills/servosity/install.ps1 | iex
Each installer drops both the CLI and the MCP server, so you can use the Skill (Claude Code / Codex) and the MCP server (Claude Desktop / ChatGPT / Cursor / etc.) from one install. For Claude Desktop, ChatGPT, Cursor, Windsurf, Cline, Continue, Gemini, or Copilot wire-up, see docs/which-agent.md and each skill's mcp-install.md.
Now what? Once your connector's
--version(ordoctor) returns clean, you're 60 seconds from your first real query. Bring your tenant + your hardest cross-client question to a free Build Session - we'll work it live with the MSP cohort and the same Skills you just installed.
Outcomes, not hype - drawn from each of the 84 connectors' skill pages:
Cross-skill questions compose too: find every ticket about backup failures across all clients by combining halopsa-cli tickets search with servosity-cli stale-backups.
The five agents MSP owners actually use:
Also works with Cursor, Windsurf, Cline, Continue.dev, Zed, GitHub Copilot, and Gemini CLI. MSP Skills speaks the open MCP standard, so any current or future MCP-capable agent can use it. Full per-tool deep-dive: docs/which-agent.md.
Run more than one agent? Install MSP Skills across all 51+ supported agents at once:
npx skills add Servosity/msp-skills@latest(requires Node.js, then run the per-skill installer for the CLI/MCP binaries). Details in docs/which-agent.md.
Most AI tools and MCP servers for PSAs and backup systems call the vendor's API on every question your agent asks. That works fine for "show me ticket #4421." It falls over at QBR time, when you're asking "how many backup-failure tickets across all 47 clients last quarter, grouped by engine?" - that's 47 paginated REST calls, rate-limit headaches, and a context window full of raw JSON the model has to re-read.
MSP Skills syncs each connected system into a local SQLite mirror with full-text search. Cross-client and cross-engine questions become one local query: instant, offline, and the AI sees the answer, not the raw data.
You don't have to switch agents to use MSP Skills. Same package, two interfaces: a Claude Code / Codex Skill for shell-style agents, and an MCP server for Claude Desktop, ChatGPT, Cursor, Windsurf, Cline, Continue.dev, Gemini, and Copilot. One install drops both binaries on your machine. Use one or both - your call. No vendor lock-in, no proprietary plugin format, no SaaS subscription that ties you to a single AI.
You don't have to know JSON, regex, or what "stdio transport" means. Paste one sentence into Claude Code or Codex and your agent reads the Skill, installs the binary, and walks you through authentication. Every command has a tier (read / write / destructive) and a recommended agent policy in each skill's governance.md - reads are always safe, and the recommended agent rule is to preview writes with --dry-run and require your approval before any mutation. The hard stuff is hidden; the safety bar is high.
These skills hold privileged, multi-tenant access to systems that run MSP businesses, so safety is a first-class concern, not a footnote:
governance.md tags commands read / write / destructive and tells you how to scope an agent. The safe default for an autonomous agent is read plus previewed writes (--dry-run first, human approval before the real thing); gate destructive and credential-touching operations behind a human.Yes, on Plus, Pro, Team, Business, Enterprise, and Education plans (Free tier does not yet expose Developer Mode). ChatGPT connects to remote MCP servers over HTTPS, not to local binaries directly. MSP Skills ships local binaries. Most of them serve Streamable HTTP themselves - run one with --transport http and put its /mcp endpoint behind an HTTPS tunnel; the six stdio-only connectors go behind a supergateway bridge instead. Step-by-step, per connector: each skill's mcp-install.md.
Yes - Claude Code, Claude Desktop, and Claude.ai web (via Claude Desktop's MCP). Claude Code reads the Skill directly. For Claude Desktop, run the installer, then go to Settings > Extensions to register the MCP server - the panel walks you through it without editing JSON. (Power users can still hand-edit claude_desktop_config.json via Settings > Developer > Edit Config if you prefer.) Both paths are first-class.
Your data stays on your machine. MSP Skills runs locally: the CLI and the MCP server are binaries on your laptop. The SQLite mirror sits in a local directory under your user account. The AI agent (Claude, ChatGPT, Codex) only sees what the CLI or MCP server returns - typically the result of a query, not raw bulk data. Credentials are read from your environment or your agent's config; they're never bundled into this repo or transmitted to MSP Skills servers (there are none). The only paths that move data anywhere else are ones you switch on and point at a URL of your own: --deliver webhook:<url>, a feedback endpoint you set, and the --transport http MCP listener on the connectors that have it.
Yes - all of them. Each speaks MCP (some natively, some via an extension or marketplace). The cross-tool table at the top of this README links each one's install path. The pattern is the same: install the MSP Skills binary, point your AI tool at it, authenticate. Tool-specific gotchas (Copilot uses servers in mcp.json not mcpServers, Cline needs an npx workaround on Windows, etc.) are documented in docs/which-agent.md.
No. The recommended install path 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, editing JSON in a text editor, or installing Node, Python, Docker, or a database. You will need to enter API credentials your PSA or backup tool gives you.
Almost never. Each connector syncs your data into a local SQLite mirror once, then incrementally; subsequent questions (triage, SLA breaches, client cards, cross-client analytics) run against the local mirror, not the live API. The big-batch QBR queries that get you 429'd with API-passthrough tools become a single SQL join here.
Vendor-native AI is great for ticket-by-ticket work - rewriting replies, summarizing one ticket, classifying sentiment. MSP Skills is the MSP-owner-on-the-couch-with-Claude layer: cross-client analytics, ad-hoc questions across thousands of tickets, multi-system queries that join your PSA with your backup tool. The two complement each other; you don't have to choose.
Yes. The PowerShell installer in "Install in 60 seconds" above is the same install path as macOS / Linux, just packaged for PowerShell. The CLI and MCP binaries are native Windows builds. Cline users on Windows may need a small npx workaround documented in docs/which-agent.md.
The Skills, the CLI binaries, and the MCP servers are free. Apache-2.0 licensed - free to use commercially, free to fork. Servosity does not charge for API access required to run the Servosity skill. Other PSA, RMM, and backup vendors set their own API-access terms. You pay only for whichever AI agent you use (Claude, ChatGPT, Codex, etc.), and those are billed by your AI provider, not by us.
Yes. Every command takes a client / company filter (e.g. halopsa-cli client card "Acme Corp", servosity-cli company show 4421). Read-tier commands change nothing; you can try them against your live system safely. For writes, each skill's governance.md tags every command with a recommended agent policy - preview with --dry-run and approve before the real thing. The safe starting move is a read-only triage against one client, see the output, then widen scope.
If your AI doesn't yet support MCP and isn't Claude Code / Codex CLI (which read SKILL.md directly), MSP Skills won't fit yet. Every connector's CLI binary still works standalone in a shell - you can pipe its output into anything. As more AI tools add MCP (the protocol is moving fast in 2026), they'll work with MSP Skills automatically without anything changing on our side.
These skills are built and tested with real MSPs running them against their own production systems, live, in our free weekly Build Sessions (Thursdays). An MSP brings a system we have not covered - your PSA, RMM, backup product, or security tool - and we co-build the Claude Code Skill and MCP server live. You watch, you ask, you walk away with a working integration, and every shipped Skill lands here the day it merges.
RSVP for the next one at compoundingteams.com/build-sessions.
We co-build the next skills live with MSPs in the weekly Build Session. The targets the MSP community asks for most, in demand order:
Want one of these next? Bring the system to a Build Session (above) or open an issue.
Easiest path - tell your AI. Paste this into Claude Code, Codex, Claude Desktop, or ChatGPT:
Open an issue on https://github.com/servosity/msp-skills using the skill-request template. The system I want is [YOUR SYSTEM NAME] - my one question for the AI would be "[YOUR QUESTION]", I'm on [Mac or Windows], and my MSP has about [N] techs. Submit it.
Your AI fills the form and posts the issue. Done. (You'll need to be signed in to GitHub in your browser or have a GitHub CLI token where your AI can see it.)
Don't have an AI set up yet? Open the issue directly: Submit a skill request →. First time filing a GitHub issue? See docs/requesting-a-skill.md for a 90-second walkthrough.
More votes (👍 reactions on an issue) move a system up the roadmap. Share with one MSP friend who'd want the same skill.
If your MSP uses a system we have not covered, send a PR. We co-build alongside contributors in Build Sessions when that is easier than going it alone.
A skill PR includes: SKILL.md (with vendor frontmatter), a README.md with the non-affiliation banner, install.sh + install.ps1, mcp-install.md if it ships an MCP server, and ideally pain-point.md + governance.md. CI enforces the contract (DCO sign-off, frontmatter schema, required files, no secrets, no personal paths). See CONTRIBUTING.md for the full checklist and the non-affiliation banner template.
MSPs are the channel that brings AI to small business. The durable moat is the Compounding Teams methodology for running an MSP where every interaction with a tool, a customer, or a system makes the next one better. Loops close, feedback returns to the source, work compounds instead of evaporating. msp-skills is the part of that methodology you can install: the executable layer that lets your AI agent operate alongside your team in the same systems, with the same context, every day.
halopsa-mcp, servosity-mcp).Standards. Conforms to the open Agent Skills spec (Anthropic, Dec 2025; 40+ agents). MCP-compatible - works with any MCP-capable agent including Hermes and OpenClaw, with metadata.openclaw frontmatter pre-wired so openclaw skills install works on every skill.
Built by Servosity. Maintained by Damien Stevens. Apache-2.0 licensed. See TRADEMARKS.md for vendor non-affiliation and SECURITY.md to report a vulnerability. Methodology: Compounding Teams. Generated CLIs and MCP servers built with CLI Printing Press.
Last updated: 2026-10-27. Latest releases: abnormal-v0.1.7 · acronis-v0.1.10 · action1-v0.1.10 · afi-v0.1.8 · appdirect-v0.1.9 · atera-v0.1.8 · autotask-v0.1.9 · auvik-v0.2.8 · avanan-v0.1.11 · aws-billing-v0.1.8 · axcient-v0.2.17 · backup-radar-v0.1.2 · betterstack-v0.1.7 · bitdefender-v0.1.4 · blumira-v0.1.6 · cipp-v0.1.13 · cisco-duo-v0.1.4 · cisco-meraki-v0.1.3 · connectsecure-v0.1.4 · connectwise-automate-v0.1.7 · connectwise-control-v0.1.8 · connectwise-manage-v0.2.4 · connectwise-rmm-v0.1.3 · cork-v0.1.8 · cove-v0.1.13 · crowdstrike-v0.1.10 · datagate-v0.1.5 · datto-bcdr-v0.1.10 · datto-rmm-v0.1.10 · domotz-v0.1.7 · fortinet-v0.1.4 · google-workspace-v0.1.3 · gradient-v0.1.8 · halopsa-v0.2.22 · hubspot-v0.2.4 · hudu-v0.1.17 · huntress-v0.1.11 · immybot-v0.1.11 · itglue-v0.1.7 · jira-service-management-v0.1.4 · kaseya-bms-v0.1.7 · knowbe4-v0.1.7 · levelio-v0.1.8 · liongard-v0.1.7 · maxio-v0.1.7 · microsoft-graph-v0.2.7 · mspbots-v0.1.6 · n-central-v0.1.8 · nerdio-v0.1.8 · ninjaone-v0.1.21 · pagerduty-v0.1.7 · pandadoc-v0.1.7 · pax8-v0.1.8 · pipedrive-v0.1.7 · proofpoint-v0.1.7 · proxmox-v0.1.2 · quickbooks-v0.1.14 · quotewerks-v0.1.3 · resourceguru-v0.1.7 · rewst-v0.1.7 · riverside-fm-v0.1.5 · rocketcyber-v0.1.7 · rootly-v0.1.7 · runzero-v0.1.7 · saas-alerts-v0.1.3 · salesbuildr-v0.1.7 · sentinelone-v0.2.6 · servosity-v0.6.11 · sherweb-v0.1.8 · skykick-v0.1.8 · sonicwall-v0.1.3 · sophos-central-v0.1.3 · superops-v0.1.13 · syncro-v0.1.7 · tactical-rmm-v0.1.7 · td-synnex-v0.1.2 · threatdown-v0.1.3 · threatlocker-v0.3.13 · unifi-network-v0.1.9 · veeam-v0.1.7 · watchguard-v0.1.3 · wordpress-v0.1.9 · xero-v0.1.9 · zammad-v0.1.10.
Every Sherweb Partner API capability, plus a local SQLite store, offline analytics, and
{ "mcpServers": { "io.github.servosity-sherweb-mcp": { "command": "mcpb", "args": [ "https://github.com/Servosity/msp-skills/releases/download/sherweb-v0.1.12/sherweb-mcp.mcpb" ], "env": { "SHERWEB_BASE_URL": "<value>", "SHERWEB_CLIENT_ID": "<secret>", "SHERWEB_CLIENT_SECRET": "<secret>", "SHERWEB_OAUTH_SCOPE": "<secret>", "SHERWEB_SUBSCRIPTION_KEY": "<secret>", "SHERWEB_TOKEN_URL": "<value>", "SHERWEB_NO_CONFIG_WRITE": "<value>", "SHERWEB_MCP_READ_ONLY": "<value>" } } }}工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式，而不是 PlugMCP 代理伺服器。
發布者列出了以下環境變數：SHERWEB_BASE_URL, SHERWEB_CLIENT_ID, SHERWEB_CLIENT_SECRET, SHERWEB_OAUTH_SCOPE, SHERWEB_SUBSCRIPTION_KEY, SHERWEB_TOKEN_URL, SHERWEB_NO_CONFIG_WRITE, SHERWEB_MCP_READ_ONLY。
Sherweb MCP 透過 stdio 在本機執行。
上次檢查時間：2026-11-15。 Every Sherweb Partner API capability, plus a local SQLite store, offline analytics, and