MCP resources & prompts: standing context and one-click analyses

Tools, resources & prompts

A first-class MCP server exposes three things. Tools are actions (covered in the data you can query). Resources are addressable, read-only context. Prompts are ready-made analyses. Your BusinessMCP endpoint advertises all three, plus server instructions describing how to use it — everything is scoped and redacted by your key's access policy.

Resources

Read these directly (no tool call) via resources/list then resources/read:

• `businessmcp://snapshot` — a live 30-day analytics + funnel + pipeline snapshot, policy-scoped.
• `businessmcp://sops` — your enabled standard operating procedures, each linking to businessmcp://sop/{id} for the full body.
• `businessmcp://memory` — the compounding knowledge the analyst has learned about your business.
• `businessmcp://brain/{query}` — hybrid search over your durable company knowledge, policy-filtered.

Prompts

Invoke by name via prompts/get. Each seeds a tool-chaining analysis:

• `weekly-growth-review` — traffic, funnel, revenue, SEO and ads, this week vs last.
• `funnel-audit` — find the biggest leak and what to fix first.
• `conversion-optimization` — CRO ideas grounded in heatmaps + replays.
• `seo-opportunities` — striking-distance keywords and quick wins.
• `ad-waste-audit` — wasted spend + audiences to build.
• `session-replay-review` — watch and summarise frustrated sessions.

Using them from a client

Most MCP clients surface resources as attachable context and prompts as slash-commands or templates. In Claude Desktop, connected resources appear in the attachment menu and prompts appear as commands. Point your client at your endpoint (see Connect Claude, GPT & Gemini) and they show up automatically alongside your tools.