01

Where AI is already useful

The useful pattern is an internal audience, a human edit before anything ships, and source data close enough to check.

  • Ticket summarization — Condense a long ticket thread before an escalation or handoff while keeping the source visible.
  • QBR narrative drafting — Turn computed service metrics into an editable executive summary without allowing a model to originate numbers.
  • Meeting prep — Summarize what changed, what was recommended, and what the client approved or declined from recorded history.
02

Where it's dangerous

The inverse pattern is an external audience, no human in the loop, or figures with no deterministic source.

  • Client-facing chat — An unsupervised bot speaking in the MSP's name can be wrong directly in front of the people who pay the MSP.
  • Invented metrics — SLA percentages, asset counts, and budget lines must be computed from source data, never generated.
  • Auto-execution — AI should suggest rather than close tickets, change configurations, or email clients without a human decision.
03

The read-only principle

Give AI read access and suggestion rights, never execution rights, and log every access. For anything a client may see, the companion rule is: numbers computed, words generated.

04

MCP, concretely

Model Context Protocol lets an AI assistant call tools exposed by other software instead of relying on pasted exports. QBR Studio's server is read-only by construction, scoped per client, and audit-logged.

  • list_clients — Returns the clients in the workspace.
  • client_metrics — Returns computed service metrics for one client and period.
  • get_report — Returns a published report's contents.
  • refresh_budget — Returns the hardware-refresh budget computed from synced asset data.
05

Where this leaves an MSP

Adopt the boring wins — summarization, drafting, and preparation — under two rules: read-only access, and numbers computed, words generated. Skip client-facing or self-executing systems until their failure mode is acceptable.

Product boundary

Keep recommendations reviewed and evidence explicit

QBR Studio computes service metrics and drafts client-facing summaries from connected data. Your MSP reviews the evidence, chooses every recommendation, and approves the final report before a client sees it.

Common questions

What MSP teams usually ask

Do MSPs actually need an 'AI strategy'?

Mostly no. They need positions on three questions: where AI drafts (internal text a human edits), where it's banned (client-facing output and anything numeric), and what access it gets (read-only). Answer those and the rest is tool selection, not strategy.

What is MCP in one paragraph?

Model Context Protocol — an open standard that lets AI assistants like Claude call tools exposed by other software. Instead of pasting exports into a chat window, the assistant queries the system directly, with the access the server chooses to grant. For MSP tooling, the interesting design decision is exactly what access that is.

Is it safe to point Claude at client data?

It depends entirely on the plumbing. Read-only, scoped, audit-logged access to computed data (the QBR Studio MCP approach) means the worst case is a wrong sentence you catch on review. Write access or raw credential sharing is a different risk class entirely. Ask any vendor: can the AI change anything, and is every access logged?

Will AI replace QBRs or the people running them?

It replaces the assembly work — pulling numbers, building decks, reconstructing history. The meeting where a human explains what the numbers mean for the client's business is the part clients are paying for, and no one has automated trust yet.