Lustereczko: open-source MCP server lets agents build live UIs mid-conversation
wuj · reddit · 2026-10-11
Lustereczko is an MIT-licensed, locally running open-source MCP server that, after five months of iteration, lets an agent build its own UI mid-conversation: the agent writes HTML that renders in chat, the app can talk back to the agent, and the agent can deploy small Python tools the app calls to do real work on your machine. The author demoed it live in GitHub Copilot in VS Code: highlight a passage of the Declaration of Independence, ask about it, and the agent answers inside the MCP app window, not the host chat.
The author argues human-LLM interaction needs rethinking and dynamic UIs should be step one. The hard parts turned out to be plumbing, not the model:
- The MCP Apps protocol only hands the app the one result that rendered it, so he added file-backed channels for two-way messages
- Some hosts strip HTML from the result, so the app renders from the tool's input instead
- Sampling (server calling the host's LLM) is deprecated and refused by the latest protocol
It started as an evening attempt to prove this couldn't be done inside MCP.
More from coding & agent
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23