Claude Code for MSPs: What You Can Automate (and Where It Falls Short)

If you’ve been using Claude Code in your MSP, you already know it’s the most capable coding tool most technicians have ever had access to. You can generate PowerShell scripts, debug API calls, refactor automation workflows, and build internal tools — all from a terminal. But “claude code msp” is one of the most searched phrases in the space right now because MSP owners and techs want to know: can Claude Code actually run my service desk? Can it triage tickets, pull device data, execute runbooks? The honest answer is that Claude Code is excellent at certain MSP tasks, good at others with the right setup, and fundamentally limited in ways that matter when you’re managing 50 clients.

This post covers what Claude Code does well for MSPs out of the box, how MCP servers extend it into your actual tool stack, what Claude skills look like in practice, and where the gap between “Claude Code on my laptop” and “production MSP automation” gets real.

What Claude Code Does Well for MSPs

Claude Code is Anthropic’s CLI tool — it runs in your terminal, has access to your filesystem, can execute shell commands, and maintains context across a conversation. For MSP work, that translates into several immediately useful capabilities.

Script generation and debugging

This is where Claude Code earns its keep on day one. Describe what you need in plain English and get working code back.

• “Write a PowerShell script that checks all users in a tenant for MFA enrollment and exports a CSV of non-compliant users” — and it writes correct Graph API calls with proper pagination, error handling, and the right $select properties.
• “This NinjaOne API script is returning 401 errors after token refresh” — paste the script in, and Claude Code identifies the OAuth flow issue, often faster than Stack Overflow.
• “Convert this ConnectWise SOAP integration to REST” — Claude Code knows the ConnectWise Manage REST API well enough to do meaningful migrations.

If your MSP runs on custom scripts — and most do — Claude Code replaces hours of Googling, testing, and debugging with minutes of conversation. For the specifics of what the ConnectWise API surface looks like for automation, we have a dedicated deep-dive.

Documentation and SOP drafting

Ask Claude Code to read a directory of scripts or config files and generate documentation for what they do, how they’re structured, and what dependencies they have. For MSPs sitting on a folder of undocumented PowerShell scripts that only one tech understands, this is genuinely valuable. It won’t replace your ITGlue articles, but it’ll draft them faster than anyone on your team will.

Internal tool development

Need a quick web dashboard for monitoring something? A CLI tool that queries multiple APIs and produces a summary? A webhook listener that parses ConnectWise callbacks? Claude Code can build these from scratch in a single session. MSPs with a technical owner or senior engineer get outsized value here — you can build internal tooling at a pace that wasn’t possible before.

Log analysis and troubleshooting

Paste in event logs, error outputs, or diagnostic dumps and Claude Code will parse them, identify patterns, and suggest next steps. For those “my computer is slow” tickets where the tech pulls a diagnostic bundle, Claude Code can analyze the output faster than most L2 engineers.

Claude Skills for MSP Work

Claude Code supports “skills” — reusable instruction sets stored as markdown files that Claude loads when relevant. For MSP work, skills let you encode your tribal knowledge so Claude Code behaves consistently.

A few examples that work well:

Ticket response skill: Instructions for how your MSP formats client communications — tone, structure, what to include, what not to say. Every tech gets the same quality of client-facing response.

Script standards skill: Your MSP’s PowerShell conventions — error handling patterns, logging requirements, naming conventions, which modules to use (Microsoft.Graph, not the deprecated AzureAD module). Claude Code follows your standards instead of generating generic code.

Triage checklist skill: When a tech pastes a ticket into Claude Code, the skill instructs Claude to ask specific diagnostic questions, suggest data to gather, and structure the response in your PSA’s internal note format.

Skills are powerful for individual productivity. The limitation is that they’re local to the machine running Claude Code. They don’t share across your team without manual distribution, and they can’t access your live tool data without MCP servers.

Connecting Claude Code to Your MSP Stack with MCP

This is where things get interesting. Claude Code supports MCP (Model Context Protocol) — the same protocol that lets Claude Desktop connect to external tools. With MCP servers configured, Claude Code can query your ConnectWise tickets, pull device data from NinjaOne, search your ITGlue documentation, and interact with M365 — all from the terminal.

You configure MCP servers in your project’s .mcp.json or via claude mcp add. Each server runs as a separate process and exposes tools that Claude Code discovers at startup. We covered the full setup — including config examples for ConnectWise, NinjaOne, ITGlue, and Microsoft Graph — in our Claude MCP servers guide for MSPs. The NinjaOne MCP server post goes deeper on the RMM-specific implementation.

With MCP configured, Claude Code becomes genuinely useful for IT support workflows:

• “What are the open P1 tickets on the Service Desk board?” — Claude Code queries ConnectWise and returns real data.
• “Check device health for ACME-DC01” — Claude Code pulls CPU, memory, disk, uptime, and patch status from NinjaOne.
• “Find the password reset SOP for Contoso” — Claude Code searches ITGlue and returns the relevant article.

This is what people searching “claude for it support” or “claude service desk” are looking for: an AI that can actually see their data and answer questions about their environment.

ChatGPT vs Claude for IT support

If you’ve tried ChatGPT for IT support — and based on search volume, plenty of MSPs have — the comparison with Claude Code comes down to two things: tool integration and code quality.

ChatGPT (including the desktop app and API) doesn’t support MCP. You can use OpenAI’s function calling to build integrations, but there’s no standardized protocol for connecting to your existing tools. You’re building custom middleware for every integration. Claude Code’s MCP support means you can connect to ConnectWise, NinjaOne, and ITGlue using servers that follow a common protocol — and the ecosystem of available servers is growing fast.

On code quality for MSP-specific tasks — PowerShell, Graph API, ConnectWise REST API — Claude consistently outperforms GPT-4o in our testing. It’s more likely to use correct endpoint paths, handle pagination properly, and avoid deprecated modules. Both models are capable, and which one is “better” changes with each model update. The structural advantage Claude has is MCP.

Where Claude Code Falls Short for MSPs

Here’s where the honesty part comes in. Claude Code is a developer tool. It’s excellent for the person sitting at the terminal. It is not a service desk automation platform, and trying to make it one will cost you more time than it saves.

No multi-tenancy

You manage 40, 80, 120 clients. When Claude Code queries ConnectWise through an MCP server, it’s using your API credentials with access to everything. There’s no built-in concept of “scope this query to Client X” or “don’t let this response include data from Client Y.” You can be careful with your prompts, but that’s not a security model — it’s a habit.

No team access

Claude Code runs on your machine. Your skills, your MCP configs, your conversation history. If five techs need Claude-assisted triage, each one needs their own setup, their own API credentials, their own skills files. There’s no shared workspace, no consistent behavior across the team, no centralized management.

No automation pipeline

Claude Code is reactive — it does what you ask, when you ask it. It doesn’t monitor your ticket queue, automatically triage incoming tickets, match runbooks, or execute resolution steps. You can use it to build scripts that do those things, but Claude Code itself isn’t watching your board at 2 AM.

No approval workflows

When Claude Code generates a script that resets a password or modifies a user account, the only approval gate is you reading the output and deciding to run it. There’s no configurable policy that says “low-risk actions auto-execute, high-risk actions require manager approval.” There’s no audit trail beyond your terminal history. For MSPs with compliance requirements, this matters.

No runbook execution

Claude Code can help you write runbooks. It can help you debug runbooks. It cannot execute a multi-step runbook that coordinates actions across ConnectWise, NinjaOne, M365, and ITGlue — checking preconditions at each step, handling failures gracefully, logging results back to the PSA, and presenting the tech with a single approval button. That’s a different category of software entirely.

We covered the five runbooks every MSP should automate first — password resets, onboarding, offboarding, device diagnostics, security alert response. Claude Code can help you build the scripts behind each one. But orchestrating them across tools, across clients, across your team, with proper guardrails? That’s the gap.

Claude Code + Junto: Where They Complement Each Other

This isn’t an either/or. The MSPs getting the most out of AI right now are using Claude Code for development work and a purpose-built platform for service desk automation.

Use Claude Code for:

• Writing and debugging scripts (PowerShell, Python, Bash)
• Building internal tools and dashboards
• Analyzing logs and diagnostic output
• Generating documentation and SOPs
• Prototyping integrations and API workflows
• Ad-hoc queries against your tools via MCP servers

Use Junto for:

• Automated ticket triage across your full client base
• Multi-tool context gathering (PSA + RMM + documentation + identity) on every ticket
• Runbook execution with configurable approval policies
• Team-wide access through ConnectWise Pod, Slack, and Teams
• Multi-tenant isolation so client data stays scoped
• 43+ pre-built runbook templates that work out of the box
• 18 AI processors that run automatically on every incoming ticket

Claude Code makes individual techs faster. Junto makes the service desk faster — whether anyone is sitting at a terminal or not.

Getting Started with Claude Code in Your MSP

If you haven’t tried Claude Code yet, start here:

1. Install it: npm install -g @anthropic-ai/claude-code (requires Node.js 18+)
2. Try a script task: Ask it to write a Graph API script you’ve been meaning to build. See how the output compares to your manual approach.
3. Add one MCP server: Connect your PSA or RMM and try querying real data. The experience of asking Claude a question and getting an answer from your actual environment is what makes the value click.
4. Build a skill: Create a markdown file with your MSP’s scripting standards or triage checklist. See how much more consistent Claude Code’s output becomes.

For MCP server setup, start with our Claude MCP servers guide. For the NinjaOne-specific implementation, see the NinjaOne MCP server walkthrough.

When you hit the wall — when you want triage on every ticket, runbooks that execute across tools, team-wide access, and multi-tenant isolation — that’s when the platform conversation starts.

---

Claude Code is great for building. Junto is built for running. See it on your actual tickets — book a 15-minute demo and we’ll show you AI triage pulling context from your ConnectWise, NinjaOne, and ITGlue data on real tickets. No MCP servers to configure.