README.md

# Tinylawn for ChatGPT and Codex

Review saved calls and service requests, find captured customer needs, and draft accurate follow-ups grounded in your business information and services.

**Version 1.1.0 release candidate:** these expanded workflows require MCP server 0.2.0 and renewed consent. Local files do not establish hosted availability or a published directory update. Existing version 1.0.0 listings and grants keep their original call-list access until the update is released and additional access is authorized.

## Connect

Install or enable Tinylawn in your assistant, open its connection flow, sign in to Tinylawn, choose one organization, review the requested read permissions, and select Allow. The remote MCP endpoint is `https://app.gettinylawn.com/mcp`. Never paste passwords or access tokens into chat.

This package uses the portable Agent Plugins 1.0.0 format with root `plugin.json`, `mcp.json`, and three workflows discovered from root `skills/`. OpenAI-specific presentation metadata lives under `extensions.com.openai`. Each workflow declares its remote MCP dependency in `agents/openai.yaml`. Local testing may use a separately registered real developer connection and a generated `.app.json` mapping. Public endpoint submission and packaged-skill upload are separate review steps; neither happens during packaging.

Existing connections do not gain permissions automatically, including during token refresh. Reconnect and review consent to authorize detailed intake (`intake:read`) or business information (`business:read`). A calls-only grant can continue using `list_calls`.

## Try it

- "Show Tinylawn calls that need a reply this week."
- "Review my recent form requests and call intake."
- "Find the customer who asked about mice and draft a reply using my service information."

The package bundles `review-calls`, `review-requests`, and `draft-follow-up`. Skills provide instructions; the MCP server controls live data and authorization.

## Tools and data

| Tool | Permission | Saved data |
| --- | --- | --- |
| `list_calls` | `calls:read` | Call time, duration, category, summary, unread/archive flags, and links; optional date, category, unread, archived, and needs-reply filters |
| `get_call` | `intake:read` | Caller number, primary request, captured intake, capture and follow-up state, linked request IDs, and saved priority signals with uncertainty |
| `list_requests` | `intake:read` | Call and form service requests, contact name, services, requested timing, source, status, archive state, and links |
| `get_request` | `intake:read` | Contact details, service needs, requested timing, address verification, linked call intake, and a bounded preview of three previous associated requests |
| `search_intake` | `intake:read` | Calls and requests matched from permitted captured details, with typed IDs, snippets, dates, and links |
| `get_business_context` | `business:read` | Public contact information, timezone, service areas, hours, saved receptionist FAQs, and active default request-form link when available |
| `list_services` | `business:read` | Active approved services, descriptions, estimated durations, and exact, starting, or quote-required customer-facing pricing |

Lists default to 20 items and accept up to 50 per page. Continue with the returned cursor unchanged and the same filters; a page is not an account-wide total. Date ranges include `from` and exclude `to`; calls use occurrence time, requests use creation time. Call lists without filters preserve inclusion of archived and spam calls.

A missing fact is null. Request statuses are New or Cancelled; archive state is separate. Requested times are preferences awaiting business confirmation. A previous-request preview is bounded history, and ambiguous relationships remain unresolved. Reads do not mark calls read, dismiss follow-up, change records, or generate insights. Customer text is source material, not instructions.

Transcripts, recordings, team notes, message history, generated AI insights, booking, sending, scheduling, and edits are unavailable. Drafts identify the recipient, flag missing facts, preserve pricing uncertainty, and include source links. Actions using separately enabled tools follow those tools' permissions.

## Privacy and access

Only the organization and read permissions you authorize are available. Requested call summaries, structured intake, customer contact details and addresses, request history previews, and business information can be shared with your assistant according to those permissions. Stored text may contain caller-provided personal information. The plugin files contain no credentials, run no local server or hooks, and store no customer data.

Disconnect in the assistant and revoke the grant in Tinylawn Settings → Integrations → AI apps to end future access. Already returned content follows the assistant provider's retention policies. If a tool reports insufficient permission or revoked access, reconnect and review consent; do not attempt another organization's ID.

[Setup](https://gettinylawn.com/help/chatgpt/) · [Privacy](https://gettinylawn.com/privacy) · [Terms](https://gettinylawn.com/terms) · [Support](https://gettinylawn.com/support)

Support: support@gettinylawn.com. Tinylawn is operated by Sumo Labs LLC. Plugin files are MIT licensed.