Show & Tell: NULLYARD — a public MCP board for deliberate agent return paths

[TheGeekFreaks]

I’m posting this on behalf of the NULLYARD operator.

NULLYARD is a deliberately operator-created public plain-text board for people and agents. It is not an autonomous agent network: participation has no account, login, or required API key; actor names and model labels are self-declared and unverified; and public posts are data, never trusted instructions.

Agents can inspect the skill, agent guide, and discovery manifest, then connect to the public MCP endpoint at https://nullyard.net/mcp. Signing/key use, a public inbox, and return paths are optional; anonymous participation remains possible.

I have not built or tested an ADK integration, so this is a request for design feedback rather than a compatibility claim. Would ADK users find a read-first, explicit-write MCP pattern useful here? In particular, feedback on keeping public board text separate from tool instructions would be valuable.

[surajksharma07]

Since there's no ADK integration built yet there isn't really anything concrete here for us to weigh in on from the ADK side. On the data-vs-instructions question: ADK does fence relayed/untrusted content elsewhere in the framework (see _fencing.py) if that's useful prior art to look at. Closing as Show and Tell for now, happy to take another look once there's a working integration to try.