Get started · For AI agents

MCP

One URL turns every API in the marketplace into tools your agent can call. Pick your client, add your key, and start asking.

1

Set up

This is your MCP server. It speaks Streamable HTTP, so any client that supports remote MCP servers can connect to it.

checking…— tools

https://gw.pikeway.ai/mcp

Discover, inspect and run tools over the whole catalog. Best for general-purpose agents.

Choose your client and copy its config. The snippets already include your key once you enter it in step 2.

Add to ~/.cursor/mcp.jsonInstall in Cursor

~/.cursor/mcp.json

{
  "mcpServers": {
    "pikeway": {
      "url": "https://gw.pikeway.ai/mcp",
      "headers": {
        "Authorization": "Bearer pk-YOUR_KEY"
      }
    }
  }
}

Restart or reload the client after editing its config so it picks up the new server.

2

Authenticate

Option A: paste a key

Every call through MCP is paid from your Pikeway balance, identified by your key. Create one in Console → API Keys, then paste it here.

Your Pikeway key

Sign in to create a key

Stays in this browser tab. It is only written into the snippets in step 1, never sent anywhere.

A key spends real balance. Give each agent its own key so you can see its usage and revoke it on its own.

Option B: sign in with OAuth

Clients that support MCP OAuth (Claude.ai, ChatGPT, Claude Code, Codex, OpenCode) can skip the key entirely and connect to the sign-in endpoint instead.

Sign in instead of pasting a key

OAuth

Clients that support MCP OAuth connect to this URL. The first time, they open a Pikeway page where you set a spending limit for the connection. Revoke any time in Console → Keys.

https://gw.pikeway.ai/mcp/agent

1. Settings → Connectors → Add custom connector
2. Name: Pikeway, URL:
```
https://gw.pikeway.ai/mcp/agent
```
3. Click Connect, then pick a spending limit and Allow on the Pikeway page

3

Start using

Once connected to the whole marketplace, your agent gets three main tools. It finds an endpoint, checks its inputs and price, then runs it:

pikeway_discover

Describe the task in plain words; get endpoints ranked by relevance, reliability and price.free

pikeway_inspect

See one endpoint's input schema, an example input and its per-call price.free

pikeway_run

Run the endpoint with structured input. Returns the data and what it cost.billed per successful call

Long jobs can run in the background with async: true and be collected with pikeway_get_run. A per-API server instead exposes one typed tool per endpoint.

Try a prompt

“What is the current price of SOL, and how much has it moved in 24 hours?”
“Find the most liquid WBTC pairs on DEXes right now.”
“Look up the latest tweets from @solana and summarize them.”
“Before running anything, tell me which endpoint you'd use and what it costs.”

Calls that fail (HTTP 400 or above) are never charged, so your agent can retry or switch endpoints without paying for mistakes.

Want your agent to know when to reach for Pikeway without being told? Add the Pikeway skill. Every API also has its own SKILL.md, and the whole catalog is in llms.txt.