Hermes Paperclip Adapter: One Tool, Zero Ambiguity

The Adapter That Actually Owns Its State

Most MCP servers are passive. They answer calls, return data, and wait. Hermes Paperclip Adapter does something more interesting — it takes ownership of status transitions, enforcing scope so your agent can't accidentally flip an issue into the wrong state with a careless prompt.

That's a small design decision with enormous practical consequences.

A Fork With Opinions

Hermes Paperclip Adapter is a MarketIntelLabs fork of the upstream hermes-paperclip-adapter originally built by Nous Research. Published to npm as @marketintellabs/hermes-paperclip-adapter, this is the version that MarketIntelLabs runs in production — not the upstream, not a prototype.

That distinction matters. When a company eats its own cooking, you get a different kind of software. Edge cases get filed. Bugs get fixed. The API stabilizes around what actually works.

The core premise: replace curl-in-prompt — the janky practice of embedding raw HTTP calls inside agent prompts — with structured, schema-validated tool calls via an in-process MCP tool server called paperclip-mcp. If you've ever watched an LLM hallucinate a curl flag and take down a workflow, you understand exactly why this matters.

One Tool, Done Right

The server exposes a single tool — and that restraint is intentional.

update_issue_status transitions an issue to one of three terminal states:

• done
• blocked
• cancelled

That's it. No freeform string. No status soup. The schema enforces an enum, which means the model cannot invent a fourth state like "kinda done" or "pending-ish". Scope enforcement is baked into the call signature itself.

This is the right way to design agent-facing tools when the downstream consequence is a persistent state change. Structured > freeform. Always.

Heartbeats and Habits

Beyond the tool itself, the adapter ships a built-in prompt template: builtin:mil-heartbeat-v3. This prompt explicitly steers the agent toward using update_issue_status rather than falling back to the old RESULT marker pattern for issue status transitions.

This is a subtle but important feature. You can expose a tool all you want — but if the agent's prompt incentivizes an older pattern, you'll get the old pattern. MIL's heartbeat prompt closes that gap by encoding the preference directly into the agent's context.

It's prompt engineering as configuration management. The tool and the prompt ship together, which means you're not left wiring them up yourself and hoping the seams hold.

By the Numbers

The server holds a composite score of 91/100 — strong across the board, with a few nuanced gaps worth understanding.

The security score of 30 is the headline here. Structured tool calls with enforced enums instead of free-text agent output is a meaningful security posture — not security theater.

The Right Audience

This adapter is not for everyone — and it's honest about that. It's purpose-built for teams running Hermes-based agent workflows who want to move away from brittle curl-in-prompt patterns without rebuilding their entire stack.

Specifically, you'll get the most out of Hermes Paperclip Adapter if you:

• Already use Nous Research's Hermes agents and want a battle-tested fork with production provenance
• Need deterministic state transitions — no hallucinated statuses, no ambiguous agent output
• Want prompt and tool to ship together rather than managing their alignment yourself
• Deploy via npm or pip — dual-ecosystem support means you're not locked into one runtime

If you're running a different agent framework, the value proposition narrows considerably. This is tightly scoped software with a clear job.

Small Star Count, Serious Engineering

Two GitHub stars. Ninety-one points. That gap tells you something about how the open-source discovery problem distorts signal.

MarketIntelLabs built this to solve a real production problem — brittle agent prompts that fail when the LLM gets creative about HTTP. They solved it the right way: structured calls, schema enforcement, and a prompt layer that reinforces the behavior you want. Then they published it.

If you're running Hermes in production and still embedding curl in prompts, this fork is the upgrade you didn't know existed.