MCP recipes
Audit stale leads in Claude: classify the five oldest and propose next steps
The assistant finds the least recently touched ENGAGED and MEETING leads, reads the five oldest conversations, and classifies each as needs follow up, needs stage correction or lost. It proposes a next step for each and changes nothing until you confirm.
Published by Sellinger AI. Numbers and settings checked on 19 October 2026 against the fact register. Report an error.
What this recipe does
A spot check, not a bulk cleanup. It reads five conversations so the findings rest on what was said, not on a record's age.
Connect the Sellinger connector first (claude.ai: Settings, Connectors, Add custom connector, paste https://app.sellinger.ai/api/v2/mcp). The steps are on the MCP page.
In claude.ai this is also available as the saved prompt audit_stale_leads, so you can start it from the prompt menu without typing.
The prompt
Audit stale leads. 1. Call list_leads for stage ENGAGED and again for stage MEETING, each ordered by updated_at ascending. 2. Call get_lead_conversation on the 5 oldest across both stages. 3. Classify each: needs follow up, needs stage correction, or lost. 4. For each, propose a specific next step with a reason. Do not send messages or change stages. Wait for my confirmation before calling update_lead or send_message.
The exact tools it calls
| Tool | What the recipe uses it for |
|---|---|
| `list_leads` | Least recently touched leads by stage |
| `get_lead_conversation` | What actually happened on each lead |
| `update_lead` | Only after you confirm: change a stage or status |
| `send_message` | Only after you confirm: send a follow up on email, LinkedIn, WhatsApp or Instagram |
The connector has 43 tools in total.
What needs your approval
- Changing a stage with update_lead and sending with send_message both wait for your confirmation.
- Do not tell the assistant to message anyone who asked to stop. The opt_out_lead tool records that.
Sample output
No sample output is published for this recipe yet. It has not been run end to end in Claude or ChatGPT for this page, and we do not invent output. The tool names and what each returns are taken from the connector's tool descriptions in the product code (checked 19 October 2026). When a real run is captured it will be added here with its date.
Notes and limits
- Updated at is an approximation of staleness. The conversation decides.
Common questions
Will it message my stale leads?
Why only five?
What about leads who opted out?
Sources and method
Tool names, parameters and warnings are from frontend/src/lib/mcp/hosted-tools.ts and the saved prompts in hosted-prompts.ts (43 registered tools, checked 19 October 2026). The recipe has not been run for this page.
Related: weekly pipeline review, outreach rules, all recipes, MCP connector.