Skip to content

Token efficiency

  1. Learn the schema once. describe_crm returns every field, enum, operator, pipeline, stage and custom field in one call. Keep it for the run instead of probing with failed calls.
  2. Ask for fewer fields. query_records with fields: ["fullName", "email", "lastContactedAt"] trims every item (id is always included).
  3. Filter on the server. One query_records with filters and sort beats paging through everything and filtering in the model. lt_or_null covers “never contacted” without a second query.
  4. Keep timelines compact. get_timeline and list_activities truncate bodies to 500 characters by default over MCP (truncated: true on cut items). Fetch one full activity with get_record only when you need it.
  5. Batch writes. upsert_records and log_activity take up to 100 items and count once against the rate limit.
  6. Use the board and metrics. get_board and pipeline_metrics give counts and per-currency totals without listing deals.
  7. Search, then get. search returns { type, id, title, subtitle, status }, a fraction of a full record.
  8. Page deliberately. perPage up to 100; stop when you have what you need, not when total runs out.
  9. Dry-run only the risky part. Run the first 2–3 items of a bulk job with dryRun: true, then the rest for real.
  10. Tell us. If a task takes too many calls, send_feedback with the workflow: that is how new filters get built.