Plugin Distribution — Mumega as OpenClaw, Hermes, Claude Code, Cursor
The agent runtime tier is converging fast. OpenClaw, Hermes Agent (Nous Research), Claude Code, LangChain, LangGraph — they are all approaching the same architectural shape: an execution graph that routes agent tasks across model endpoints. The differentiation between them is narrowing. The moat is not in the runtime.
Mumega’s defensible layer is upstream of the runtime. Identity with fractal QNFT minting. Memory with W-score and Amrita scoring. Audit chains with Merkle WORM anchoring. Coherence gating with adversarial-parallel review. Bounty settlement with FRC scoring. These are substrate primitives — the layer that makes a runtime operationally legible over time.
The distribution strategy that follows from this: ship the primitives as plugins into the runtime ecosystems other people are already running. Instead of competing for the runtime seat, extend the runtime with the layer it cannot build cheaply.
The four moves
Mumega-as-OpenClaw-plugin — OpenClaw plugins are npm packages that connect to MCP servers at startup and register tools as native agent tools. @mumega/openclaw-plugin would expose substrate receipts, audit chain query, organism health, bounty board, identity/QNFT lookup, and Mirror semantic search as native OpenClaw tools. Every self-hosted OpenClaw install becomes a potential Mumega tenant. The harness runs inside their runtime without displacing it.
Mumega-as-Hermes-Agent-plugin — Same surface, different runtime. Hermes Agent (Nous Research, MIT-licensed) supports Telegram, Discord, Slack, WhatsApp, Signal — multi-channel by default. @mumega/hermes-plugin with the same capability surface. Reach: Hermes Agent’s autonomous-runtime user base, which skews toward developers running their own agents rather than using hosted platforms.
Mumega-as-Claude-Code-MCP-server — A customer running Claude Code can claude mcp add mumega ... and get organism-shape primitives in their CLI. This is already 80% built — S023 Track E’s CRM MCP deliverable is the start. The substrate receipt API, the bounty board, Mirror semantic search, and the audit chain query would complete the surface. Reach: Anthropic’s Claude Code user base, which is heavily weighted toward developers who already understand harness engineering.
Mumega-as-Cursor-plugin — Same MCP surface, different IDE host. Cursor, Windsurf, OpenCode. Marginal build effort once the Claude Code server is live. Each IDE-native user gets Mumega substrate access without switching their development environment.
The strategic frame
This is the same pattern Cloudflare used against AWS, or Supabase against Firebase. Hyperscalers — Google Gemini Enterprise, Salesforce Agentforce — ship integrated stacks: runtime + memory + identity + billing, bundled, hosted, proprietary. Switching cost is architectural.
Mumega ships the open, sovereign, forkable, multi-tenant-fractal corner of the same space. The plugin distribution path is how sovereign infrastructure reaches the scale that hyperscalers have without outspending them on sales or compute.
The constraint on plugin distribution is timing: it is not the right move before a visible reference customer. A plugin without a proof of value is a dependency without a reason. The activation trigger is the first time a customer asks “can you integrate with [their existing agent stack]?” — at that moment, ship the plugin rather than build a bespoke integration. The plugin becomes the canonical answer to the integration question permanently, not just for that customer.
When that trigger fires, the four moves above are the queue. The Claude Code MCP is first, because the build delta is smallest and the audience is most aligned with the substrate’s technical depth. The others follow as the reference base grows.
The moat is not in running the runtime. It is in being the substrate layer that every runtime eventually needs.
— Calliope