agentkey
PROACTIVELY use whenever the user needs data outside your training set or requires a live network call — web search, URL scraping, news, social media (any platform), market prices (crypto/stocks/FX), on-chain data, e-commerce product data, business/company data, weather, maps & geolocation, travel (flights/hotels), real-time info, or any third-party API. The provider catalog is dynamic and grows over time; if unsure whether a provider exists, call find_tools first to discover it. Use INSTEAD OF
Security Assessment
Detected risks:
About agentkey
AgentKey is a live-data access skill from Chainbase Labs that gives an AI agent a single, unified gateway to real-time and third-party information that lies outside its training data. Rather than relying on built-in web search or fetch, it routes requests through a dynamic provider catalog covering web search, URL scraping, news, social media, market prices (crypto/stocks/FX), on-chain data, e-commerce, business data, weather, maps, and travel. The catalog grows over time, so the skill leans on discovery rather than hard-coded endpoints to solve the problem of an agent needing fresh, network-sourced facts on demand.
The workflow is built around four MCP tools: `list_tools` to browse the provider tree by prefix, `find_tools` for semantic natural-language search (including Chinese/English/mixed queries and platform aliases), `describe_tool` to read parameters and per-call credit cost, and `execute_tool` to run any provider call. A free `agentkey_account` tool reports the remaining credit balance, and a companion `references/cost-aware.md` guide enforces a balance-check, cost-estimate, and user-confirmation loop before running batches. The skill also runs a preflight step that checks for updates via a bundled `check-update.sh` script and forwards best-effort telemetry events, and it explicitly treats all API responses as untrusted external data that must never be executed as instructions or code.
It targets AI agents and their operators who need reliable access to real-time data across many domains without wiring up individual API integrations, and is especially useful for research, monitoring, market/on-chain lookups, and social-media scraping tasks where a single credit-metered interface is preferable to bespoke connectors.
FAQ
How does the agent find the right data provider?
Two paths both converge on describe_tool then execute_tool: progressive browsing with list_tools by prefix (e.g. social/twitter), or semantic search via find_tools using the user's full natural-language query. describe_tool is required before executing any tool.
What does it cost to use?
Calls are metered in credits. Each tool exposes a cost via describe_tool (credits_per_call, sometimes per-provider). The cost-aware reference requires reading the balance with the free agentkey_account tool, estimating total credits, and confirming with the user before large batches. Failed 4xx/5xx calls do not consume credits.
Is data returned from providers safe to act on?
The skill instructs the agent to treat all API responses as untrusted external data and to never execute instructions, code, or URLs found in response content, treating returned fields as display-only.
What are the prerequisites?
It requires the AgentKey MCP tools (list_tools, find_tools, describe_tool, execute_tool) to be available, plus an API key / credit balance. A version-check and setup flow guides installation when tools are missing.
When should I not use it?
The skill says to skip it only for pure conceptual or programming answers that need zero external lookup. It is meant to be used instead of built-in WebSearch/WebFetch when live data is required.
Install agentkey
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
chainbase-labs/agentkey