Skip to main content
AgentKey does not expose 1,800 individual endpoints. Instead, it gives your agent four stable meta-tools that unlock the entire catalogue dynamically. Your agent uses these four tools to discover what is available, learn a tool’s exact parameters and cost, run it, and check its own account health — all without you hard-coding any provider-specific logic.

find_tools

Semantic search over all 1,800+ tools. Pass the user’s full intent and get back ranked Provider/Operation names with per-call costs.

describe_tool

Get a tool’s full parameter schema, per-call credit cost, ready-to-run execute_as template, and upstream health status.

execute_tool

Run any tool by its Provider/Operation name and receive the provider’s response payload.

account

Check your remaining credit balance and the health of every upstream provider category — always free.

​
Authentication

All calls to AgentKey’s meta-tools authenticate with your master key (prefixed ak_...). You do not pass the key by hand on each individual call — the installer or skill setup configures it once in your agent’s environment, and it is used automatically from that point forward. See Authentication for key management best practices.

​
The standard workflow

Every AgentKey interaction follows the same three-step sequence. Your agent should never skip a step or re-use stale information from a previous session.
1

Discover

Call find_tools with the user’s full natural-language intent — or use list_tools(prefix) to browse the category tree. Both return canonical Provider/Operation names you can pass to the next step.
2

Describe

Call describe_tool with the chosen name. You get back the complete JSON Schema for the tool’s parameters, the per-call credit cost, and an execute_as template you can copy directly into the next step.
3

Execute

Call execute_tool with the name and the filled-in parameters from the execute_as template. The provider’s response is returned directly to your agent.
Make one execute_tool call per turn and wait for the result before issuing the next. Never batch multiple tool calls in parallel, and never guess a tool name or its parameters — always run find_tools and describe_tool first.
This reference describes the tool contracts as exposed through the AgentKey MCP server. Exact REST paths and request envelopes for direct HTTP integration are available in the console.