Back to stack
Layer 03/ 05

Claude.

Writes the brief, scores ICP fit, wired into the record via MCP.

Claude sits right on the HubSpot record via MCP. The prompts are built per use case and run in batch, and every token lands as a log in BigQuery — you can watch along.

Batch
whole CRM list, one run
<2min
per account brief
Cents
per enrichment, right in the token log
All
runs logged in BigQuery
How I solve it

How I set it up for you.

  • Prompts versioned in Git, deployed via CI
  • Batch enrichment over the entire CRM list
  • MCP server feeds live records to the model
  • Token cost transparent per workflow + per lead
Toolchain
Anthropic APIMCP ServerVercel AI SDKPrompts-as-CodeTool CallsStreamingPrompt CachingToken LogsAnthropic APIMCP ServerVercel AI SDKPrompts-as-CodeTool CallsStreamingPrompt CachingToken Logs
Record → MCP → Claude → brief
Record
Acme GmbH
42 emp · DE
Live fetch
MCP Server
crm.read
Model
Claude
claude-sonnet-4.51,248 tok
Enriched
ICP-Match
0.87+ brief.md

MCP server fetches the live HubSpot record, Claude analyzes, scores, writes the brief. Token cost transparent.

Example workflow

Example: ICP-match enrichment

  1. 01Account ID arrives from HubSpot
  2. 02MCP fetches properties + LinkedIn snapshot
  3. 03Claude analyzes and scores ICP fit
  4. 04Score + brief land back in HubSpot

Want me to build this for you?

30 min demo. I walk you through a real setup, live.