Use the MCP server#

MCP lets an AI client (Claude, Microsoft 365 Copilot, or anything else that speaks the protocol) call Reportworq directly: list content, run a job, fetch output, ask about a report. Timberline is almost entirely pre-configured for it, with one deliberate exception.

The one switch#

AllowMcpAccess is off in the shipped demo store. Nothing MCP works until an administrator turns it on in the authentication settings.

That is the safe default for a shared workspace, but it means MCP needs one deliberate setup step before you can use it.

What is already configured#

So once access is on, both channels, impersonation, auditing and sane link limits are already in place.

Note the two demo conveniences that are not recommendations: public download links and a never-expiring authenticated-link setting. Say so if the room is security-minded. The point is that these are per-deployment choices, and yours would be different.

Setting up a demo#

1. Sign in as an administrator and enable MCP access in the authentication settings.
2. Pick a channel. Direct is the simpler story on a laptop; CloudHub is the one to show if the conversation is about a hosted deployment.
3. Connect your client. The mechanics differ per client. See the customer AI & Integrations guide for the client-side setup, which this guide does not duplicate.
4. Verify with something read-only first: list the workspace content, and confirm the eight jobs and two campaigns come back.
5. Check the MCP audit afterwards and show the calls that were made. It is a good closing beat for a governance-minded audience.

What to actually ask it#

Point questions at content that was built to answer them.

The answer-layer report templates that ship with the demo exist largely for this. Six parameterized workbooks (sales trend, income statement, margin mix, balance sheet and cash flow, and sales ops), each with:

• an AI Context tab describing what it is and is not for,
• a Control tab with named parameter cells (ReportYear, MarketSel, VersionSel),
• an embedded data tab tagged with a geography key, so one formula scopes to Total Company, a region or a market,
• report tabs whose columns run Jan Feb Mar Q1 … Q4, FY, prior year, YoY%.

They are granular and self-describing, which is exactly what an MCP question needs. Ratios are always recomputed rather than summed, so a margin question gets a correct answer rather than an averaged one.

The jobs themselves carry AI profiles, so questions about which report should I use have something to work with. See Check out the AI features.

Good demo questions:

• "What drove sales in February, and by which product line?" Sales Information answers this well.
• "What was FY2026 gross margin for EMEA?" A scoped answer from the answer layer.
• "Run the Financial Statements job for North America and give me the PDF." The run-and-fetch path, which is the one people actually want to see.
• "Which report should I use for a specific invoice?" This should route you away, because the profiles say so.

Acting as another user#

With impersonation enabled, an administrator can exercise MCP as a specific user without collecting passwords. Use it to show that AI access is scoped: ask the same question as a Leadership user and as an APAC user, and show the difference.

This is a testing and demonstration facility. Treat it as an administrator capability, and mention the audit log in the same breath: impersonated calls are recorded.

Before the meeting#

• Turn MCP access on, and check it is on. This is the step people forget.
• Confirm the AI provider connection is valid: MCP and the in-app AI both depend on it.
• Confirm the job's data source is reachable if you plan to run a job rather than just list content.
• Decide whether you are demoing the direct or CloudHub channel, and set the client up in advance.

Related#

• Check out the AI features
• Customer guide: AI & Integrations: Copilot & MCP