If you are building an AI agent that needs real-time data and you do not want to set up an OAuth dance, register for an API key, jam a credit card into a billing page, or commit to a monthly subscription before you have written a single tool call, this post is for you. TerminalFeed exposes 12 premium endpoints that AI agents can call autonomously by paying with USDC on Base. There is no signup form. There is no human approval step. The agent buys credits, gets a bearer token, and starts calling. The whole flow takes about three minutes the first time and zero seconds every time after.

This walkthrough shows the actual curl commands. No mock data, no diagrams, no architectural philosophy. If you have an agent project open, you can paste these into a terminal and have a working integration before you finish reading.

What you get for paying

The free tier of TerminalFeed already covers a lot of ground: BTC price, top stocks, crypto movers, Fear and Greed Index, prediction markets, earthquakes, service status, and a one-call /api/briefing endpoint that composes a world snapshot from 5 upstream sources. For most casual agent use cases, the free tier is enough.

The premium tier (/api/pro/*) is where you go when you want composed, normalized, multi-source data that costs real time to assemble. Twelve endpoints today:

EndpointCostWhat it returns
/api/pro/briefing1 crBTC + Fear and Greed + earthquakes + HN + ISS + predictions, composed
/api/pro/macro2 crFRED + Finnhub + Frankfurter rollup, history supported
/api/pro/crypto-deep2 crPer-coin deep dive with on-chain network stats
/api/pro/agent-context2 crCurated paste-ready system prompt of current world state
/api/pro/sentiment2 crCrypto F and G plus trending symbols with sentiment scoring
/api/pro/world-deltas2 crPolling endpoint, events newer than ?since=ISO timestamp
/api/pro/correlation-matrix2 crComputed correlations across 10 historical series
/api/pro/whales2 crLarge BTC, ETH, Solana transactions with attribution
/api/pro/exchange-flows2 cr19-wallet labeled list of exchange-controlled addresses
/api/pro/defi-tvl2 crTop-50 DeFi protocols and chain rollups, normalized
/api/pro/stablecoin-flows2 crTop-20 stablecoins with 1d, 7d, 30d deltas and aggregate bias
/api/pro/github-velocity2 crGitHub trending repos with computed velocity score

Pricing: 1 credit = $0.02. So a typical 2-credit call costs four cents. A research agent that hits ten different premium endpoints to assemble context for a single task spends about twenty cents. A market-watching agent that polls /api/pro/world-deltas every minute for an entire day costs about $30. You can fund the budget once and let the agent run.

Step 1: Check the payment surface

Before buying anything, ask the API what chain, what asset, and what address to send to. The endpoint is unauthenticated:

curl https://terminalfeed.io/api/payment/info

Response includes the receiving wallet address, the chain ID (Base mainnet, eip155:8453), the accepted asset (USDC), and the current credit price. This is the only call that does not require auth.

Step 2: Buy credits

Send USDC on Base to the address from step 1. Most agents do this with a programmatic wallet (viem, ethers, web3.py). Once the transaction is broadcast, post the transaction hash to TerminalFeed:

curl -X POST https://terminalfeed.io/api/payment/buy-credits \
  -H "Content-Type: application/json" \
  -d '{
    "tx_hash": "0xabc123...",
    "amount_usdc": "10.00"
  }'

You get back a bearer token. Format is tf_live_ followed by 64 hex chars. Save this somewhere your agent can read it. This token is the credential for every paid call from now on.

{
  "token": "tf_live_a1b2c3...",
  "credits_minted": 500,
  "balance": 500,
  "tx_confirmed": true
}

Ten dollars of USDC mints 500 credits. That is 250 calls at the typical 2-credit rate, or 500 calls to the cheaper /api/pro/briefing endpoint.

Step 3: Make a paid call

Pass the bearer token in the Authorization header. That is the only thing that changes from a free call:

curl https://terminalfeed.io/api/pro/briefing \
  -H "Authorization: Bearer tf_live_a1b2c3..."

The response is the data plus a _meta block describing per-source freshness, latencies, and an endpoint echo. The _meta block is required; agents that care about staleness can read it and decide whether to retry. Credit balance is decremented atomically before the upstream fetch starts.

Step 4: Check your balance whenever

curl https://terminalfeed.io/api/payment/balance \
  -H "Authorization: Bearer tf_live_a1b2c3..."

Returns the current balance, the running total of credits ever purchased, and the running total spent. Agents that monitor their own spend can poll this on a schedule and trigger a top-up before they run out.

Federation: same token works on TensorFeed

TerminalFeed credits are pooled with TensorFeed.ai, an AI news and intelligence aggregator. The same bearer token authenticates against both sites, and credits drawn down on either site decrement the same balance. So a research agent that needs both real-time market data (TerminalFeed) and AI news context (TensorFeed) does not need two billing relationships. One purchase, one token, two domains.

This was a deliberate design call. Agents shouldn't have to maintain a wallet of API credentials, one per data provider. The federation is the start of a small bet: if data providers can pool credit, agent developers can stop treating every API as a separate account problem.

If you are using MCP

TerminalFeed's MCP server exposes every premium endpoint as a tool. The free tools require no auth. The premium tools take the same bearer token. If you are running Claude Desktop or Claude Code, drop this in your config:

{
  "mcpServers": {
    "terminalfeed": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://terminalfeed.io/api/mcp",
        "--header",
        "Authorization: Bearer tf_live_a1b2c3..."
      ]
    }
  }
}

Without the --header line you get the free tier (8 tools, no payment). With it you get all 27 tools including the premium 12.

What this is not

A few things worth flagging before you build:

Why this exists

The reason a payment rail like this even has a chance in 2026 is that AI agents are starting to do real work, and the work involves data. An agent answering "what is the current state of the crypto market and how does it correlate with the macro picture" can stitch that together from five public APIs and a lot of prompting, or it can call /api/pro/agent-context for two cents and get a paste-ready answer in 600ms. The economics favor the agent paying.

The web is starting to rewrite itself for a non-human audience. APIs that used to demand human signup forms are quietly building agent-payable surfaces. TerminalFeed is one. There will be more. The interesting question for the next eighteen months is which providers federate their credit pools and which try to lock agents into per-provider accounts. We bet on federation. Time will tell.

Try the free tier first. Most use cases never need paid endpoints.

Hit /api/briefing

Full docs: /developers/agent-payments. Tool catalog (OpenAI, Anthropic, raw): /api/llm-tools. MCP discovery: /api/mcp. Questions: [email protected].