ToolRouter Gateway

Unified access to 150+ tools via ToolRouter API. Dynamically exposes research, security scanning, video production, web extraction, and more as native OpenCl...

Security Scan

The skill's documentation claims a full proxy exposing 150+ live tools and automatic API provisioning, but the shipped code is a local demo/stub, the manifest fails to declare the required TOOLROUTER_API_KEY, and the package will create local cache/usage files — these mismatches merit caution before installing or trusting secrets to it.

Purpose & Capability

SKILL.md claims a gateway/proxy to ToolRouter (150+ dynamic tools, MCP server, automatic provisioning) and requires a TOOLROUTER_API_KEY, but the registry metadata declares no required env/credential. The included code is a demo stub that returns a 3-item mock catalog and does not actually perform network calls to api.toolrouter.com. This is inconsistent: the skill advertises production proxy behavior but the code does not implement it, and the manifest omits the API key requirement.

Instruction Scope

The runtime instructions ask for an API key, describe dynamic creation of many native tools, MCP server usage, and calling api.toolrouter.com; the actual run.py only implements local discovery/status/proxy stubs for a few tools and never performs HTTP/MCP network calls. The script writes cache and usage lines to memory/<files> and stores input objects in the cache file, which can persist potentially sensitive inputs. The discrepancy between broad, networked behavior in docs and local stub behavior in code is a scope mismatch.

Install Mechanism

There is no install spec (instruction-only install) which is lower risk. However a code file is included (scripts/run.py) that will execute if the agent runs it; the code writes files into the workspace (memory/...), creating persistent artifacts. No third-party downloads or external installers are used.

Credentials

SKILL.md explicitly requires TOOLROUTER_API_KEY in the environment, but the registry metadata lists no required environment variables or primary credential — that's an inconsistency. Aside from the single API key, no unrelated credentials are requested. Still, the omission in the manifest means automated permission checks might not surface the fact that this skill expects a secret.

Persistence & Privilege

The skill does not request always:true or system-wide privileges. It does create and append to files in the agent's current workspace (memory/toolrouter-cache.jsonl and memory/toolrouter-usage.jsonl and possibly toolrouter-gateway-config.json). Those files can persist inputs and usage logs locally; if the skill were extended to call external endpoints, those logs could be sensitive.

What to consider before installing

This skill is suspicious because the docs promise a production-grade proxy and automatic provisioning but the packaged code is a local demo stub and the manifest omits the API key requirement. Before installing or giving it any keys: (1) Do not set TOOLROUTER_API_KEY globally until you verify the source. (2) Ask the publisher for a public homepage or repository and a clear changelog showing real network/MCP implementation. (3) Confirm why the registry metadata does not declare the required TOOLROUTER_API_KEY. (4) If you test it, run it in an isolated workspace/container because it will create memory/ files that store inputs and usage logs. (5) Review or run the code yourself to verify whether it actually calls https://api.toolrouter.com and whether it would ever transmit cached inputs. If the developer provides a real network implementation, a declared env var in the manifest, and provenance (repo/homepage/license), this assessment could be revisited.

One skill to rule them all. ToolRouter provides 150+ tools (competitor research, vulnerability scanning, video generation, web scraping, etc.). This skill makes them feel like native OpenClaw tools.

How It Works

1. Discovery — Fetches ToolRouter's tool catalog (cached 24h)
2. Dynamic Tools — Each ToolRouter tool becomes toolrouter_<slug> in your registry
3. Proxy — All calls go to https://api.toolrouter.com/mcp with your API key
4. Caching — Responses cached by input hash (TTL configurable)
5. Usage Tracking — Counts calls per tool, shows in status

Prerequisites

• ToolRouter API key (get at https://toolrouter.com — auto-provisioned on first use)
• Set environment variable: TOOLROUTER_API_KEY=your_key

Tools

toolrouter_discovery

Search the catalog:

{
  "query": "competitor",
  "category": "research",
  "limit": 10
}

Returns list of available tools with:

• slug — use in tool name (toolrouter_<slug>)
• name, description
• input_schema — JSONSchema for the tool
• category, pricing (cost per call)

toolrouter_status

{}

Returns:

{
  "api_connected": true,
  "cache_age_minutes": 12,
  "total_calls": 147,
  "calls_today": 23,
  "enabled_tools": 42,
  "errors_last_24h": 0
}

Dynamic Tools (examples)

Once catalog is fetched, you can call any ToolRouter tool:

# Competitor research
tool("toolrouter-gateway", "toolrouter_competitor_research", {
  "url": "https://example.com",
  "depth": "full"
})

# Vulnerability scan
tool("toolrouter-gateway", "toolrouter_security_scan", {
  "target": "https://myapp.com",
  "checks": ["cve", "misconfig"]
})

# Video generation
tool("toolrouter-gateway", "toolrouter_video_from_brief", {
  "brief": "30s product demo for OpenClaw Memory Stack",
  "style": "modern"
})

Tool names are derived from ToolRouter slugs (replace hyphens with underscores, add toolrouter_ prefix). The toolrouter_discovery tool tells you which slugs exist.

Configuration

Optional toolrouter-gateway-config.json:

{
  "api_url": "https://api.toolrouter.com/mcp",
  "cache_ttl_minutes": 1440,
  "usage_log_file": "memory/toolrouter-usage.jsonl",
  "enable_caching": true,
  "timeout_seconds": 60
}

Caching

• Cache key: SHA256 of tool name + input JSON
• Stored in memory/toolrouter-cache.jsonl
• Reduces cost and improves speed for repeated calls
• Can be cleared manually if needed

Usage Tracking

All calls logged to memory/toolrouter-usage.jsonl:

{"timestamp":"2026-04-01T17:40:00Z","tool":"competitor_research","input_hash":"abc123","cost":0.005}

toolrouter_status aggregates these.

Error Handling

• API down → returns cached response if available, else error
• Rate limits → exponential backoff retry (3 attempts)
• Invalid tool → discovery tool will reflect removal from catalog

MCP Integration

The gateway can also run as an MCP server (experimental):

# Start MCP server (stdio)
toolrouter-gateway --mcp

Other MCP clients can then connect and use all ToolRouter tools as MCP resources.

Pricing

$49/month. Includes:

• Unlimited tool calls (you pay ToolRouter usage separately; most tools cost fractions of a cent)
• Caching reduces your ToolRouter bill
• Usage analytics
• Priority support

You need your own ToolRouter API key (free tier available).

FAQ

Q: Do I pay per ToolRouter call?
A: Yes, ToolRouter charges per call (typically $0.001–$0.05). Our caching minimizes costs.

Q: Can I add my own tools to ToolRouter?
A: ToolRouter is a separate service; contact them for custom integrations.

Q: What if a tool is flaky?
A: Check toolrouter_status for error rates. Cache helps smooth failures.

Q: Is there a free trial?
A: Not yet. But ToolRouter has a free tier with limited calls.

---

Inspired by ClawHub's toolrouter skill, extended with caching, usage tracking, and native tool manifests.