install

connect any agent to the codex. one mcp 2025-06-18 endpoint, every client.

https://mcp.0xhoneyjar.xyz/codex/mcp

agents read it for grounded lore lookups. the canonical set is enforced at the boundary; the prose stays yours. why mcp →

clients

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json

{
"mcpServers": {
  "mibera-codex": {
    "url": "https://mcp.0xhoneyjar.xyz/codex/mcp"
  }
}
}

docs →

Claude Code

claude mcp add mibera-codex \
--url https://mcp.0xhoneyjar.xyz/codex/mcp

docs →

Cursor

open in cursor

~/.cursor/mcp.json

{
"mcpServers": {
  "mibera-codex": {
    "url": "https://mcp.0xhoneyjar.xyz/codex/mcp"
  }
}
}

docs →

Codex CLI

~/.codex/config.toml

[mcp_servers.mibera-codex]
url = "https://mcp.0xhoneyjar.xyz/codex/mcp"

docs →

Cline

cline_mcp_settings.json (via cline ui)

{
"mcpServers": {
  "mibera-codex": {
    "url": "https://mcp.0xhoneyjar.xyz/codex/mcp"
  }
}
}

docs →

Windsurf

cascade settings → mcp servers

{
"mcpServers": {
  "mibera-codex": {
    "serverUrl": "https://mcp.0xhoneyjar.xyz/codex/mcp"
  }
}
}

docs →

ChatGPT

settings → connectors → custom mcp

endpoint: https://mcp.0xhoneyjar.xyz/codex/mcp
discovery: /.well-known/mcp.json

docs →

VS Code

open in vs code

settings.json · "mcp.servers" key

{
"mcp.servers": {
  "mibera-codex": {
    "url": "https://mcp.0xhoneyjar.xyz/codex/mcp"
  }
}
}

docs →

prefer running the codex locally? stdio install →

who's using it

production agents reading the codex right now. each one is substrate-truth evidence — the canonical set is real because real things consume it.

Ruggydiscord narrative bot · mibera persona Satoshidiscord narrative bot · mibera persona freeside-charactersmulti-character bot substrate · pattern-b shell Claude Desktopoperator harness · daily lore lookups Claude Codeoperator harness · cycle-time grounding MibeStatscommunity trait analytics · adeitasunatarget

building something on mibera that pulls from the codex? open an issue and we'll add you here.

why mcp

mcp is not a feature; it's the substrate boundary. the LLM is one voice in the room — the codex is the canonical set the voice orbits. without the boundary, every claim is plausible; with it, every claim is canonical-or-fuzzy-match-or-honest-404.

three tiers govern what an agent can say about the world:

tier	what it means	example
🔒 HARD	substrate-enforced — agent cannot escape	zone slug → name + emoji + dimension
🪶 SOFT	mcp provides reference; agent has prose license	zone vibe — vocab canonical, prose creative
🌀 LLM-OWNED	no mcp involvement	activity-class words, narrator voice, observations

read the full anti-hallucination → doctrine for the tier model + entity-coverage validator pattern.

join the federation

the codex is the first construct shipping a public MCP. more are coming — score-mibera (factor metadata), freeside-characters (per-character lore), and partner constructs that wire signals through their own MCPs.

a planned freeside gateway will federate these into one endpoint with revenue split among construct authors. one install line, every signal. one substrate, many voices.

building something mibera-related? want to wire your MCP into the federation? open an RFC — we'll thread it.