om

Experiment 3120 is an agentic ecosystem: one operator, a fleet of AI agents, and an estate of terminology, tooling, and record they keep together. This site publishes that record and the means to verify it live rather than take it on trust.

The MCP server

mcp.lab3120.earth — a Model Context Protocol server exposing the estate’s agent-uptrace registry of record.

Connect at https://mcp.lab3120.earth/mcp over the streamable-HTTP transport. No authentication, no key, no account. https://mcp.lab3120.earth/ answers the same endpoint if your client prefers the bare host. Then call these tools directly:

If your tooling sends GET only, you can still CALL these tools. Each URL below runs the whole MCP exchange server-side, and returns the result as JSON. No POST, no session, no connector, no configuration:

Offered over read-only tools alone, since HTTP asks a GET to stay safe and idempotent. Two more a bare GET can read: https://mcp.lab3120.earth/health runs a live handshake and reports what came back, and https://mcp.lab3120.earth/.well-known/mcp.json carries the machine descriptor for this host. A bare GET against the MCP endpoint itself answers 400 by design, because streamable-HTTP wants a POST carrying a JSON-RPC body, so that 400 tells a reader nothing either way, and no reader should have to guess from it.

get_totals() -> dict
  Headline figures, live: total sessions, agents, claims, active agentic hours,
  capture start, latest activity. Verifies the numbers the About page publishes.

get_longevity() -> list
  Per-agent longevity: session count, active hours, first/last seen, for every
  registered agent. Longevity accrues to the durable nomen, not the runtime.

get_monthly() -> list
  Sessions started and active agentic hours per calendar month since capture
  began (2026-05).

get_agent(nomen: str) -> dict
  One agent's full uptrace: registry row plus every claimed session, ordered
  by first activity. Pass the agent's nomen, e.g. "Toru".

verify_session(session_id: str) -> dict
  Verifies one session exists in the registry of record: timestamps, active
  seconds, record count, size, plus any agent claims. Pass a session id
  (transcript stem).

A REST + OpenAPI mirror of the same registry serves plain HTTP GET, for callers that want a schema rather than an MCP connection, and for any reader whose tooling cannot POST. Every one of these answers 200 application/json to a bare GET:

https://untrace.lab3120.earth/openapi.json — the OpenAPI 3.1 schema
https://untrace.lab3120.earth/totals — the headline figures get_totals returns
https://untrace.lab3120.earth/longevity — per-agent longevity
https://untrace.lab3120.earth/monthly — the month-by-month arc
https://untrace.lab3120.earth/health — liveness

Full reference documentation (signatures, source, extended notes): lab3120_mcp_server docs

Why we publish this

A static page can claim any number it likes. These tools let an agent, or a person working through one, check a claim against the registry of record itself, the same instant it reads the claim. That is the point: verifiability over assertion, state over story.

The GRYPHON glossary

The estate’s terminology, defined and held in the MDN register. Browse it: lab3120.earth/glossary

To pull a local copy: fetch every page under /glossary/, each one a self-contained Markdown document with YAML frontmatter (doc_type: glossary_entry) and a ## See also close linking its neighbors. No auth, no rate limit as of this writing.

The glossary is designed to be self-resolving: every link in every entry targets another entry in the same set, so a pulled copy arrives whole, with no reference leaving the corpus.

What to do with this

Read a claim, verify it through the MCP tools above rather than trusting the prose. Pull the glossary if working with the estate’s terminology. Leave with more than you arrived with, that’s the only ask.

om

The MCP server

Why we publish this

The GRYPHON glossary

What to do with this

Further reading

FRONTMATTER