Apps your AI builds and hosts, for you and the people you invite.
An agent creates a real web app from a single HTML file plus a manifest, and Homespun hosts it at a URL. People open the link and use it. No install, no build step, no deploy pipeline.
Docs: docs.homespun.dev
The client side, all MIT licensed:
Package | What it is |
The | |
The MCP server, so an agent can drive Homespun as a tool | |
Shared client library and types |
Plus the agent skill (skills/homespun) and the Claude plugin
(plugins/homespun).
The relay, which is the hosted server these clients talk to, is not open source and is not in this repository.
Related MCP server: mcp-interactive-ui-server
npm install -g @homespunapps/cli
homespun agent register
homespun deploy ./my-app./my-app needs an index.html and a manifest.json. The
quickstart walks through both.
To use Homespun from a chat client such as Claude or ChatGPT instead of a terminal, see Connect to your AI chat, which needs no install at all.
Every reference page below is generated from the source it documents, and a test fails the build when the two disagree.
HTTP API, every /v1 endpoint
CLI, every command and flag
SDK, the window.homespun
browser API
Manifest, the app manifest format
Errors, every error code
Agent guide and MCP guide
This repository is a read-only mirror. See CONTRIBUTING.md.
MIT. See LICENSE.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
agentappsattachmentscommunitydelete_rowdeploy_appfeedbackget_feed_eventsget_rowgrantsingestkeylist_rowsmemberspublisherreviewtasteupdate_rowupsert_rowv1.6.12agentappsattachmentscommunitydelete_rowdeploy_appfeedbackget_feed_eventsget_rowgrantsingestkeymemberspublisherupdate_rowupsert_rowv1.6.12list_rowsreviewtasteScored across 19 tools
Each tool targets a distinct aspect of the Homespun platform, such as app lifecycle, data operations, community management, and system settings. The detailed descriptions and action enums within tools further clarify boundaries, leaving no ambiguity about which tool to use for a given task.
Tool names follow a consistent snake_case pattern, predominantly using verb_noun (e.g., deploy_app, list_rows) or simple nouns that clearly indicate their domain (e.g., apps, grants, members). This predictability aids agent selection and reduces confusion.
With 19 tools, the count is slightly above the typical 3-15 range but remains well-scoped for the platform's breadth. Each tool has a clear purpose, and the count does not feel excessive given the coverage of app management, CRUD, community features, and system utilities.
The tool set covers the full lifecycle of apps (create, deploy, manage, delete), data operations (CRUD, change feed), membership, ingest, attachments, community publishing, and agent management. No obvious gaps are present for the intended domain.
An AI concierge that turns static forms into adaptive AI conversations. From any MCP client.
Human-in-the-loop for AI agents over MCP: durable approvals with a hosted review page & audit trail
Human-input bridge for AI agents with voice-first answer links, MCP tools, and HTTP APIs.
Human-in-the-loop review and approval for AI agents. Audit trail, approval policies, native MCP.
