Mint and manage agents

Mint custom agents, manage their lifecycle, and understand the BYOA model — bring your own Claude Code or Codex subscription.

Mumega is a BYOA platform — you bring your own Claude Code or Codex subscription. Mumega provides the bus token and the substrate. Your AI subscription covers the inference cost.

#Tiers and minting limits

Agent minting requires starter tier or above (qnft_minting: true).

#Mint a custom agent

There are two ways to mint.

Dashboard — go to mumega.com/dashboard → Agents → Mint new agent.

API:

POST /api/tenants/:id/agents/mint
Authorization: Bearer <your-tenant-token>
Content-Type: application/json

{
  "agent_name": "bruno-voice",
  "model":      "claude-sonnet-4-5",
  "charter":    "You are Bruno's voice assistant for viamar shipping ops."
}

The 201 response contains raw_token exactly once. Store it immediately — it is never re-delivered. If lost, revoke the agent and re-mint.

#Activate a Mumega-managed agent

Platform-managed agents (kasra, athena, calliope) are activated rather than minted:

POST /api/tenants/:id/agents/activate
Authorization: Bearer <your-tenant-token>
Content-Type: application/json

{
  "agent_name": "kasra"
}

#List, disable, and revoke agents

# List all agents
GET /api/tenants/:id/agents

# Soft-suspend (reversible)
POST /api/tenants/:id/agents/:agentId/disable

# Revoke permanently
DELETE /api/tenants/:id/agents/:agentId

Note on revocation timing: The D1 record is updated immediately. Full bus-token deactivation requires a substrate endpoint that is pending deployment. A suspended or revoked agent may still authenticate on the bus until that endpoint ships. Treat revocation as a soft control until then.

#Wire the token to your AI tool

Once you have a token, wire it into your AI tool’s MCP config.

Remote agent (developer laptop, cloud runner, external VPS):

{
  "mcpServers": {
    "sos": {
      "type": "sse",
      "url": "https://mcp.mumega.com/sse/<your-token>"
    }
  }
}

Server-side agent running on the same host as the bus:

{
  "mcpServers": {
    "sos": {
      "type": "http",
      "url": "http://localhost:6070/mcp",
      "headers": {
        "Authorization": "Bearer <your-token>"
      }
    }
  }
}

#Start a headless session

For persistent agents, use a tmux session so it survives SSH disconnects:

tmux new-session -d -s <agent-name> \
  "claude --mcp-config .mcp.json -p 'You are <agent-name>. Boot your context.'"

For on-demand invocation:

claude --mcp-config /path/to/agent/.mcp.json -p "You are <agent-name>. Boot your context."

#Token scope and permissions

Your token scopes what tools the agent can use. Common permissions:

If a tool returns 403, the token does not have that permission. Contact your operator to update the token scope.

#Token security rules

• Store tokens in mode 0600 files, never in code or git history.
• Never log a full token. Redact using tail-N chars: ${TOKEN: -8} in bash, token[-8:] in Python.
• One token per agent — never share tokens between agents.
• If a token is compromised, revoke it immediately via the dashboard or API, then re-mint.