FundraiseUp · MCP server — Interactive MVP

Grounded in [SPEC] MCP server + apps/mcp source

What Brandon asked — and where to see it

A clickable walk-through of the MCP server: the dashboard surface where an org turns MCP on, and the agent-side flow showing prompt → tool → internal service → response → formatted answer, for the read-only tools that exist today.

Sample surfaces the user engages with (dashboard only?)

For v1, yes — a Dashboard setting to enable/disable MCP per organization (org-admin only), plus the client-connect snippet.

Sample prompts / plain-text requests customers would enter

One natural-language prompt per tool — click any to run it.

Which tools should be called

Each prompt maps to one of the real read-only tools registered in apps/mcp; the tool is shown as a badge in the run.

Which APIs the MCP server references

None public. Each tool calls an internal apps/mcp service (e.g. TotalRaisedService, PerformanceService) over FRU internal data — the MCP does not use the public API.

Sample datasets returned

Each run shows the structured payload with the real output-schema field names from the tool source.

How the response should be formatted

Tools answer in product language (no raw field/method names); the agent renders a short summary + table.

How to use this prototype

Toggle MCP off on the Dashboard tab, then run any prompt on the Agent tab — you'll see the exact failure behavior from the spec (MCP access disabled for this company). The state pill in the header reflects it everywhere.

Dashboard → Settings → MCP connectors

Model Context Protocol (MCP)

Let AI agents (Claude Desktop, ChatGPT, custom agents) query this organization's data in natural language, inheriting each user's dashboard permissions.

Enable MCP access for this organization

Enabled — connected clients can call the read-only tools below.

Enabled

While disabled, every MCP request for this organization is rejected with:

MCP access disabled for this company

🔒 Only an organization administrator can change this setting. Members with settings access can view the current state.

Available tools (read-only)

Connect a client

Available now

Point any MCP-compatible client at the FRU remote server. (v1, on staging, authenticates via the dashboard session.)

Tools inherit your dashboard permissions and are scoped to your companyId.

Agent experience

MCP enabled

Pick a plain-text request a fundraiser might type. Each run shows the full pipeline: tool called → internal service → sample response → formatted answer. (Answers Brandon's points 2–6.)

Tip: turn MCP off on the Dashboard tab to see the disabled-state behavior.

Tools & services reference

The read-only tools registered in apps/mcp, the internal service each calls, the data scope and PII handling. The MCP is a standalone service that reads FRU internal data — it does not call the public API.

Tool	What it does (from source)	Internal service (apps/mcp)	Scope & PII

All tools are read-only and scoped to the caller's companyId, inheriting the user's dashboard permissions. Write tools (create/update campaigns, designations, copy generation) are out of scope for v1. Per spec, sensitive/PII handling (masking, no-PII mode) and payment-data exclusion are target requirements for the platform version.