Multi-Client MCP Setup: How One Agency Manages 30 Accounts in Claude

The multi-client problem agencies actually have

If you run a marketing agency, the standard tooling assumes single-client setups. Supermetrics charges per data source per client. AgencyAnalytics requires you to log in and out per client. Even most MCP servers built in 2026 are single-account by default — you connect one Google Ads, one Meta, you query, you log out, you connect the next.

Real agency workflow is the opposite. You manage 10-30 clients simultaneously. The questions you ask change all day:

• "What's the CPL trend across all my clients?"
• "Pull last 30 days for client A's Meta Ads."
• "Switch to client B's GA4 — which page lost the most traffic this week?"
• "Compare ROAS for clients C, D, and E. Which has the best return on Meta vs Google?"

The right MCP setup makes these one-prompt queries instead of three-tool tab-switches.

Auth architecture for multi-client work

The core requirement: one OAuth grant per platform per client account. You connect your agency's master account once. Each client account gets its own OAuth handshake to Google Ads, Meta Ads, GA4, and Stripe. Every connection is scoped to that one client's data — no possibility of cross-contamination.

In 1ClickReport, the model is:

• One agency-level account ($25-99/mo subscription, just like a solo founder)
• Multiple "workspace" or "client" memberships within that account
• Per-client OAuth scopes — connecting Client A's Google Ads grants access only to that client's account, never bleeds to Client B

The implication: your monthly cost stays flat ($25 or $99) regardless of how many clients you connect. Compare this to Supermetrics' per-connector pricing or AgencyAnalytics' per-client tier pricing, and the math gets interesting fast.

The account switcher pattern

The most-used workflow in agency mode is the explicit account switch. You tell Claude: "Switch to client [name]'s context for the next set of questions" — then ask questions normally without re-specifying the client each time.

Example session:

You: List my client accounts.
Claude: [returns 14 clients]

You: Switch to MyPedia Clinic's context.
Claude: Active context: MyPedia Clinic. Connected: Google Ads, GA4, Meta Ads.

You: What's the CPL trend for the last 30 days?
Claude: [pulls data only for MyPedia Clinic]

You: Now switch to Solomon People Solutions.
Claude: Active context: Solomon. Connected: Google Ads, Meta Ads, Search Console.

You: Same question — CPL trend last 30 days.
Claude: [pulls Solomon's data only]

The MCP server tracks the active context per conversation. You never have to re-auth. You never see Client A's data when asking about Client B. Mistakes are functionally impossible because the data layer enforces the scoping.

Cross-client queries (the real time-saver)

Once you have 10+ clients connected, the most valuable workflow becomes cross-client comparison. You're not asking about one client — you're asking about your portfolio.

Examples we hear from agencies running this stack:

• "Which 3 of my 14 clients had the worst Meta ROAS week-over-week?"
• "List clients whose Google Ads spend is up but conversions are flat — these need an audit call this week."
• "For all clients running Performance Max, what's the average asset group CTR?"
• "Which clients haven't used a specific feature (Advantage+, Smart Bidding, etc) that we should pitch?"

Each of these would take 60-90 minutes in a traditional dashboard tool, switching between accounts, exporting CSVs, building pivot tables. In a properly-set-up MCP agency stack, it's one prompt and 15-30 seconds.

That's the real ROI of multi-client MCP. Not "the dashboard is prettier" — it's "the 30-minute weekly portfolio review becomes a 4-minute scan."

Billing and permissions for team setups

Two questions agencies ask before they commit to a multi-client MCP setup:

1. Can my team members each have their own access?

Yes — but with care. Inviting 5 team members to a shared MCP workspace means 5 people can run cross-client queries simultaneously. For agencies that want strict per-client access (e.g., the account manager for Client A can only see Client A), workspace-level permissions handle this.

Most small agencies skip this complexity and run a shared workspace where all team members can see all clients. For 5-20 client agencies this is fine. For 50+ client agencies, per-client permissioning becomes necessary.

2. How do I bill clients for MCP usage?

Most agency MCP tools (including ours) don't pass through usage costs to clients — you pay a flat subscription, you absorb the cost in your retainer. This is the same model as Supermetrics, AgencyAnalytics, etc.

If you want to track per-client usage for internal margin analysis or value-based pricing, the MCP server's logging (per-query, per-client-context) gives you the data. Most agencies don't bother — the savings are large enough that detailed cost allocation isn't worth the operational overhead.

An anonymized real example

One agency we work with through 1ClickReport runs 30+ active client accounts across SaaS, ecommerce, and B2B service industries. Their old stack: AgencyAnalytics for client-facing dashboards, Supermetrics for data connectors into Google Sheets, manual exports for monthly reviews.

Total old stack cost: roughly $1,400/month combined across AgencyAnalytics, Supermetrics, and additional GSheets storage. Plus 8-12 hours/week of an analyst's time on report assembly.

What they replaced it with:

• 1ClickReport Pro ($25/mo) for the MCP layer + 30 client connections
• AgencyAnalytics retained for client-facing dashboards (because clients want to log in and see numbers)
• Supermetrics + GSheets entirely removed

The analyst time dropped from 8-12 hours to 1-2 hours per week. The "what should I focus on this Monday" portfolio scan became a 5-minute Claude session instead of a Sunday-night spreadsheet exercise. Combined savings ~$600/mo in tooling plus ~30 hours/month of analyst time.

This pattern is what we see most often in agency MCP setups: MCP replaces the analyst-time-heavy tools, not the client-facing tools. Both layers stay; one gets dramatically cheaper.

Frequently Asked Questions