MCP workflows

Examples of what to ask an assistant connected to Leadveyor, and what it does behind the scenes. Phrase requests your own way — the assistant picks the tools from their descriptions.

Daily review

What came in since yesterday? Which hot leads have no outcome yet?

  1. search_leads with tier: "hot", outcome: "none" — compact list with scores and summaries.
  2. list_inbounds with classification: "junk" — what was filtered out, with the reason for each.
  3. For anything that looks interesting, get_lead — the full lead with every criterion's answer.

A read key is enough.

Why did this lead score like that?

Lead lead_2Ph6Tc9Ks4Wm1Lx7Qv3Nb8Dr got 35 — why so low?

The assistant calls get_lead and reads the evaluation: every criterion with its result (met, not_met, unknown), the points it gave, and the exact quote from the message. It can then tell you that, for example, the budget wasn't mentioned (unknown, 0 points) and the event is 90 km away (rule not met). If the lead was scored again after a profile change, history: true shows the earlier evaluations.

Record outcomes

Jane Doe's wedding is booked, $18,500. The Smith kitchen went to a competitor on price.

The assistant finds each lead with search_leads or manage_contacts (action: "list", email: …), then calls record_outcome — won with amount: 18500, lost with reason: "Competitor, price". Outcomes are what show later which criteria really predict a sale.

Tune a profile

We've started doing corporate events and they're very profitable. Score them high, and test it on these three emails before changing anything.

  1. list_profiles, then get_profile — reads the current criteria and points.
  2. generate_criterion with the description — gets a criterion with a question, guidance and levels with points.
  3. update_profile_draft with upsertCriteria — adds it to the draft. Live scoring doesn't change yet.
  4. test_profile with the three emails — shows tier and score for each, with the new criterion's answer.
  5. The assistant shows you the results and asks. Only after your yes: publish_profile.

If the new version scores worse in practice: "Roll back the profile to the previous version" → get_profile with versions: true, then rollback_profile.

Try a message without saving it

How would this score? "Hi, we need catering for 40 people next Friday in Bondi, budget $2k."

evaluate_lead with the text. Nothing is stored and no webhook fires — safe to try anything. Ask "with the draft" to use test_profile instead and see the effect of unpublished changes.

Set up a new business

Set up Leadveyor for a plumbing business in Brisbane: emergency jobs and bathroom renovations are the best leads, we don't do commercial work. Give me an email address to forward enquiries to and send hot leads to https://hooks.example.com/leads.

  1. manage_workspaces (update) — business description, address, time zone and currency (needs an admin key).
  2. list_templates → create_profile with templateId: "plumbing" and variables: { "service_radius_km": 30 } — a draft with job types, emergencies escalated to hot, service area, timing and contact criteria.
  3. update_profile_draft — adjusts it to what you said: more points for bathroom renovations, a disqualify criterion for commercial work. Without a fitting template the assistant uses generate_profile with your description instead, and shows you its assumptions.
  4. test_profile on a few typical enquiries, then publish_profile after you agree.
  5. manage_sources (create, type: "email") — returns the forwarding address.
  6. manage_destinations (create) with filter on evaluation.tier = hot — returns the signing secret; test sends a test event.

A lead was marked junk by mistake

The message from bob@example.com yesterday was a real enquiry, not spam.

  1. list_inbounds with classification: "junk" — finds the message.
  2. reprocess_inbound with notJunk: true — processes it again, skipping the junk filter.
  3. get_processing_status — confirms the lead was created and scored.

To stop it happening again, the assistant can add the sender's domain to the profile's junk.allowDomains with update_profile_draft (set), then publish.

Check webhook delivery

Did my CRM get today's leads?

manage_destinations with action: "deliveries" shows each delivery's status, attempts and the CRM's response. A failed one can be sent again with action: "retry"; a destination disabled after repeated failures is turned back on with update, status: "active".

Tips

  • Start read-only. Give the assistant a read key until you trust its workflow, then a write key.
  • Use a test key to experiment. Profiles, sources and leads created with a test key stay out of live data.
  • Ask for a test before publishing. Profile changes affect every new lead; test_profile costs a few cents and shows the effect first.
  • Be specific about time. search_leads and list_inbounds return newest first; "since yesterday" works because the assistant pages until it reaches older items.

Was this page helpful?