MCP developer prompts¶
These prompts are written for developers using CorpusIQ as an MCP server — either from Claude Desktop, a custom MCP client, or an agent pipeline. They go beyond operator-level queries and focus on programmatic access, cross-connector joins, debugging, and building CorpusIQ into larger workflows.
If you're new to MCP, see how-it-works/mcp-architecture.md.
Querying business data programmatically¶
"List all tools available for the Shopify connector."¶
MCP call: resolve_connector with intent "list Shopify tools"
What this does: Returns the exact tool schemas callable against
Shopify. Useful as a discovery step before building a pipeline that
needs specific Shopify fields.
Developer use case: Enumerate available actions before generating
tool-call code in your agent.
"Fetch all orders from the last 30 days and return them as structured JSON."¶
Connectors used: Shopify or Amazon Seller.
MCP tool: get_orders with created_after date param.
What this does: Returns raw order objects with line items, amounts,
statuses. The MCP layer returns typed JSON — no parsing of prose needed.
Developer use case: Feed order data into a downstream analytics
pipeline or a fine-tuning dataset.
"Pull GA4 sessions by channel for the last 90 days, grouped by date."¶
Connectors used: GA4.
MCP tool: run_report with dimensions ["date", "sessionDefaultChannelGroup"]
and metric ["sessions"].
What this does: Returns a per-day, per-channel session matrix.
Developer use case: Build a channel attribution table without a
separate BI tool.
"Return QuickBooks P&L for Q1 as structured data."¶
Connectors used: QuickBooks.
MCP tool: get_profit_loss with start_date and end_date.
What this does: Returns income, cost of goods, expenses, and net
profit as a structured report object.
Developer use case: Ingest into a spreadsheet, dashboard, or
financial model automatically.
"Get all active HubSpot deals with their amounts and pipeline stage."¶
Connectors used: HubSpot.
MCP tool: list_hubspot_deals with pagination.
What this does: Returns deal ID, name, amount, stage, and owner
for every open deal.
Developer use case: Build a pipeline forecast tool that refreshes
from live CRM data.
"List every Klaviyo campaign sent in the last 60 days with open and click rates."¶
Connectors used: Klaviyo.
MCP tool: get_campaigns with start_date and end_date.
What this does: Returns campaign objects with recipient count, open
rate, click rate, and revenue attribution.
Developer use case: Automate an email performance report that runs
on a schedule.
Cross-connector joins¶
"Join Shopify order revenue to Meta Ads spend by date and compute ROAS."¶
Connectors used: Shopify, Meta Ads.
MCP tools: get_orders (Shopify) + get_account_insights (Meta Ads)
with matching date ranges.
What this does: Two separate MCP calls return parallel date-keyed
datasets. Your agent joins on date and computes ROAS = revenue / spend.
Developer use case: Build a custom attribution model that doesn't
rely on Meta's self-reported ROAS.
"Compare GA4 sessions on email send days to non-send days."¶
Connectors used: Klaviyo, GA4.
MCP tool (built-in cross-source): correlate_email_vs_web_traffic.
What this does: Returns a day-level table: email sent (bool), GA4
sessions, delta from baseline.
Developer use case: Validate email's halo effect on organic traffic
before attributing it as a channel.
"Join QuickBooks invoices to HubSpot deals to find deals that closed but were never invoiced."¶
Connectors used: HubSpot, QuickBooks.
MCP tools: list_hubspot_deals (filter: stage=closed-won) +
list_invoices (QuickBooks).
What this does: Two lists, joined on company name or amount. Gaps
are deals with no matching invoice.
Developer use case: Revenue leakage audit — catch won deals that
slipped through billing.
"Pull every Stripe charge and match to QuickBooks income lines."¶
Connectors used: Stripe, QuickBooks.
MCP tools: list_charges (Stripe) + get_profit_loss or
list_invoices (QuickBooks).
What this does: Stripe returns raw charges; QuickBooks returns
booked income. Mismatches surface timing gaps or missing entries.
Developer use case: Automated reconciliation check — run daily in
a CI pipeline.
"Show me Klaviyo list growth alongside Facebook ad spend to see if paid ads are growing my list."¶
Connectors used: Klaviyo, Meta Ads.
MCP tool (built-in cross-source): correlate_form_signups_vs_ad_traffic.
What this does: Returns form submit counts and Meta spend per day.
Developer use case: Validate paid-to-list growth attribution.
Debugging connector issues¶
"Why is my Shopify connector returning zero orders?"¶
Connectors used: Shopify.
Debug approach: Call get_orders with an explicit wide date range
(created_after = 90 days ago). If still zero, call
get_connector_status and check for token expiry or scope issues.
Common causes: Token revoked in Shopify admin, wrong store domain
configured, date filter is too narrow by default.
"My GA4 report is returning empty rows — how do I debug it?"¶
Connectors used: GA4.
Debug approach: First call list_accounts to verify the property
is visible. Then call list_properties with the account ID. Then run a
minimal report: single dimension date, single metric sessions, last
7 days. If that works, add dimensions one at a time until the empty
response reappears.
Common cause: property_id has the format properties/XXXXXXX —
make sure you're passing just the numeric ID, not the full path.
"Check whether my QuickBooks token is still valid."¶
MCP tool: get_company_info — this is the lightest QuickBooks
call. A valid token returns company name and fiscal year. An expired
token returns a 401 error, which tells you to re-authenticate via
get_connector_status.
"List all connectors and their current auth status."¶
MCP tool: get_connector_status (no parameters).
What this does: Returns a table of every configured connector,
its auth state (connected / expired / not authenticated), and a
re-auth link where applicable.
Developer use case: Health check endpoint in an agent — run this
at session start to gate which connectors are usable.
"Why are my Meta Ads impressions different between the account summary and the campaign-level report?"¶
Connectors used: Meta Ads.
Debug approach: get_account_insights aggregates across all
campaigns including deleted/archived. list_campaigns with status_filter=ALL
then get_campaign_insights per campaign will surface whether deleted
campaigns are inflating the account total.
Fix: Filter campaigns by status_filter=ACTIVE for apples-to-apples.
Building agent workflows with CorpusIQ as the knowledge layer¶
"Build an agent that answers CFO questions using live QuickBooks + Stripe data."¶
Architecture:
1. Connect Claude (or your LLM) to CorpusIQ MCP at https://mcp2.corpusiq.io/mcp.
2. Authenticate QuickBooks and Stripe connectors.
3. The agent calls get_profit_loss, get_balance_sheet, list_charges,
and get_balance as needed, grounded in live data.
No ETL required. CorpusIQ handles the OAuth tokens, rate limiting,
and data normalization.
"Build a daily business digest that fires at 8am and pulls from 5 connectors."¶
Architecture:
1. Cron job (or scheduler) sends a prompt to your MCP-connected LLM at 8am.
2. Prompt: "Run the executive snapshot skill."
3. CorpusIQ's executive-snapshot skill pulls QuickBooks, Shopify,
HubSpot, ad accounts, and GA4 automatically.
4. The LLM returns a formatted digest. Push to Slack, email, or Notion.
MCP skill trigger: invoke_skill with skill_name="executive-snapshot".
"Build a Slack bot that answers revenue questions."¶
Architecture:
1. Slack bot receives a message like "how did we do last week?"
2. Bot forwards the message to an LLM with CorpusIQ MCP attached.
3. CorpusIQ calls Shopify + QuickBooks, returns structured revenue data.
4. LLM composes a Slack-formatted reply. Bot posts it.
Key connector: get_profit_loss + get_orders for revenue.
Read-only guarantee: CorpusIQ cannot post to Slack, only read.
"Use CorpusIQ as the memory layer for a sales agent — give it access to all open deals and customer history."¶
Architecture:
1. Attach CorpusIQ MCP to your sales agent's LLM context.
2. On each call, the agent resolves which connectors to query via
resolve_connector.
3. list_hubspot_deals, get_hubspot_contact, list_invoices
(QuickBooks), and list_gmail_messages give the agent a 360-degree
view of each account.
Why MCP beats RAG here: Data is live, not stale embeddings. The
agent always sees the current deal stage, latest invoice status, and
most recent email thread.
"Expose CorpusIQ data to a custom dashboard without a backend."¶
Architecture:
1. Use any MCP client library (e.g. @modelcontextprotocol/client-typescript).
2. Connect to https://mcp2.corpusiq.io/mcp with your CorpusIQ JWT.
3. Call tools directly from your frontend or edge function.
4. Render the returned JSON in your dashboard components.
No separate API to build. CorpusIQ's MCP layer is the API.
"Run a competitive SEO audit on demand from a CI pipeline."¶
Connectors used: Semrush, Search Console, GA4.
MCP tools: get_domain_overview, get_competitors, get_performance.
Architecture:
1. CI job calls your MCP-connected LLM with: "Run an SEO health check
and flag any position drops > 5 places since last week."
2. CorpusIQ pulls live Semrush + GSC data, compares to prior run
(stored in canonical facts), and returns a diff report.
3. CI job posts the report to Slack or opens a GitHub issue on regressions.