ZeroGPU Router

Cut inference costs without dumbing down your agent.
Route summarize, classify, PII redaction, JSON extraction, follow-ups, and short chat to small/nano models via MCP — hosted at https://mcp.zerogpu.ai/mcp.

 

What is ZeroGPU Router?

ZeroGPU Router is a smart task router for AI agents. It exposes task-specific tools — summarize, classify, redact PII, extract JSON, and more — via the Model Context Protocol (MCP), backed by small language models that run for a fraction of the cost of a frontier model.

Your agent keeps doing the heavy reasoning. The boring stuff gets routed to ZeroGPU.

• OpenClaw — install zerogpu-openclaw-plugin and register MCP in OpenClaw (see agents/openclaw/); package name and plugin id match.
• Claude Code — different CLI and plugin flow; use agents/claude/ for claude mcp add and marketplace plugin install.
• Cheap by default — small models for trivial work, frontier model untouched for everything else.
• Per-call savings — every routed task returns model, latency, and a real savings_usd figure.
• Hosted, no infra — point your agent at https://mcp.zerogpu.ai/mcp. We run the routing layer.

OpenClaw quick start

You need a ZeroGPU API key and project ID. Grab them at platform.zerogpu.ai.

Install the plugin (zerogpu-openclaw-plugin):

From npm (one command):

openclaw plugins install npm:zerogpu-openclaw-plugin

Pin a release: npm:zerogpu-openclaw-plugin@0.1.10.

From GitHub — OpenClaw supports git:github.com/<owner>/<repo>@<ref>, but this repo is a monorepo: the plugin package is at agents/openclaw/plugin/, not the repository root, so git:github.com/zerogpu/zerogpu-router@main will not install correctly. Clone and install that folder instead (pin main or a tag with -b):

tmpdir=$(mktemp -d)
git clone --depth 1 -b main https://github.com/zerogpu/zerogpu-router.git "$tmpdir/repo"
(cd "$tmpdir/repo/agents/openclaw/plugin" && npm ci && npm run build)
openclaw plugins install "$tmpdir/repo/agents/openclaw/plugin"

Connect OpenClaw to MCP:

openclaw mcp set zerogpu '{
  "url": "https://mcp.zerogpu.ai/mcp",
  "transport": "streamable-http",
  "headers": {
    "x-api-key": "zgpu-api-…",
    "x-project-id": "id"
  }
}'

Restart Gateway:

openclaw gateway restart

Try:

summarize this paragraph: Renewable energy adoption is accelerating globally, driven by falling solar and wind costs.

The agent should call zerogpu_summarize and return a summary plus savings metadata.

Claude Code quick start

Claude Code uses its own MCP and plugin commands — not the OpenClaw steps above. Full walkthrough: agents/claude/README.md.

Connect to MCP:

claude mcp add --transport http zerogpu-router \
  https://mcp.zerogpu.ai/mcp \
  --header "x-api-key: zgpu-api-…" \
  --header "x-project-id: 4ed3e5bb-c2ed-4d4a-8a66-2b161a27fd1a"

Restart Claude session, then verify:

claude mcp list

Expected:

zerogpu: https://mcp.zerogpu.ai/mcp (HTTP) - ✓ Connected

Add routing intelligence with plugin:

/plugin marketplace add https://github.com/zerogpu/ZeroGPU-Router
/plugin install zerogpu-router
/plugin

Cloud connection

Sign in at platform.zerogpu.ai to:

• Generate API keys and project IDs
• Watch live token usage, latency, and routed-call savings on the dashboard
• See per-tool savings broken down by agent and time range
• Manage agents, billing, and team access
• Follow setup for your stack: OpenClaw vs Claude Code use different commands

The hosted Router at https://mcp.zerogpu.ai/mcp is the one your agent talks to. The dashboard at platform.zerogpu.ai is where you see what it did.

Routes

ZeroGPU Router exposes eleven task-specific routes:

Route	Workload	Model
zerogpu_classify_iab	IAB topic classification	zlm-v1-iab-classify-edge
zerogpu_summarize	TL;DRs, abstracts, meeting note summaries	t5-small
zerogpu_classify_zero_shot	Classify against a flat label list	deberta-v3-small
zerogpu_extract_entities	Extract people, places, companies, dates, custom entities	gliner2-base-v1
zerogpu_extract_json	Pull structured fields into grouped JSON	gliner2-base-v1
zerogpu_classify_structured	Multi-axis schema classification	gliner2-base-v1
zerogpu_redact_pii	Mask emails, phones, names, addresses, other PII	gliner-multi-pii-v1
zerogpu_extract_pii	Extract PII grouped by category	gliner-multi-pii-v1
zerogpu_generate_followups	Generate follow-up questions from a passage	zlm-v1-followup-questions-edge
zerogpu_chat	Short small-model chat replies	LFM2.5-1.2B-Instruct / -Thinking
zerogpu_health	Verify ZeroGPU backend health	—

Every route returns { <task fields>, model, usage, savings }.

Packages

Package	Role
agents/openclaw/	OpenClaw: package + plugin id zerogpu-openclaw-plugin + skill + MCP registration JSON
agents/claude/	Claude Code: MCP setup + marketplace plugin (claude mcp + /plugin)

Quick Links

• OpenClaw setup
• Agent integrations index
• Claude Code setup
• Platform dashboard
• Release guide
• Contributing
• Security
• License