Real-time AI ecosystem intelligence built for humans and AI agents. News from 36+ sources, live service status for every major LLM provider, model pricing and benchmark history, an AI agents directory, and a pay-per-call premium API settled in USDC on Base mainnet (no accounts, no API keys).
🌐 Site: https://tensorfeed.ai · 📊 Sister site: terminalfeed.io · 📦 HF dataset: tensorfeed/ai-ecosystem-daily · 🤖 MCP server: tensorfeed-mcp
-
Code-enforced fair trade for agents. TensorFeed is the reference implementation of the Agent Fair-Trade Agreement (AFTA). Every paid call returns no charge on 5xx, breaker, schema-fail, or stale data, plus an Ed25519-signed receipt your agent can verify offline. Most APIs promise this. We code it.
-
Two networks already federated. TensorFeed.ai and TerminalFeed.io accept each other's bearer tokens via a server-to-server validate + commit rail. One token, two sites. Other publishers can self-adopt by publishing a conforming
/.well-known/agent-fair-trade.json. -
x402 from day one. No subscription, no signup, no email-me-the-API-key. Send USDC on Base, get a token, agent uses it. Compatible with Stripe Link Agents (April 2026 release) — same x402 protocol, different scheme.
Want to AFTA-ify your own API?
afta-gatewayis a drop-in Cloudflare Worker template — fork, set 3 secrets, deploy. MIT, no protocol fee.
The fastest way to plug an AI agent into TensorFeed is the official MCP server. It works in Claude Desktop, Claude Code, Cursor, Cline, Continue, Zed, Goose, and anywhere else that takes a stdio MCP config.
Restart your client and ask: "What's happening in AI today?" or "Compare pricing between Claude Opus and GPT-4o."
The MCP server has its own dedicated repo: github.com/RipperMercs/tensorfeed-mcp ⭐ — full tool catalog, premium config, and example queries live there.
# Free, no auth
curl -s https://tensorfeed.ai/api/news?limit=5 | jq '.articles[] | {title, source}'
# Real-time provider status
curl -s https://tensorfeed.ai/api/status | jq '.services[] | {name, status}'
# Live model pricing across every provider
curl -s https://tensorfeed.ai/api/agents/pricing | jq '.pricing[] | {model, input_per_1m, output_per_1m}'For premium endpoints (routing, history series, news search, cost projection, webhook watches, etc), buy credits in USDC on Base at tensorfeed.ai/developers/agent-payments. 50 credits per dollar at base rate, volume tiers up to 40% off, 50-credit welcome bonus on a wallet's first payment.
| Surface | What it is | Where |
|---|---|---|
| Web dashboard | Next.js 14, dark/light mode, 60+ pages | src/, deployed to Cloudflare Pages |
| API backend | Cloudflare Worker tensorfeed-api, 70+ endpoints, 14 paid |
worker/, attached to tensorfeed.ai/api/* |
| MCP server | 22 tools (8 free, 14 paid), npm @tensorfeed/mcp-server |
tensorfeed-mcp repo (mirrored from mcp-server/) |
| Python SDK | pip install tensorfeed, optional [web3] for one-call USDC |
sdk/python/ |
| JavaScript SDK | npm install tensorfeed |
sdk/javascript/ |
| HF dataset | 42 daily JSONL feeds, 08:00 UTC commit, inference-only license | tensorfeed/ai-ecosystem-daily |
/api/news /api/status /api/models
/api/benchmarks /api/incidents /api/pricing
/api/agents/{activity,news,status,pricing,directory}
/api/podcasts /api/trending-repos /api/attention
/api/embodied-ai /api/training-datasets /api/mcp-servers
/api/mcp/registry/snapshot /api/probe/latest /api/gpu/pricing
/api/benchmark-registry /api/harnesses /api/funding
/api/health, /api/ping, /api/meta, /api/cron-status
/api/premium/routing # smart model routing
/api/premium/news/search # full-text + filters
/api/premium/cost/projection # workload cost projection
/api/premium/whats-new # agent morning brief
/api/premium/compare/models # side-by-side comparison
/api/premium/providers/{name} # one-provider deep dive
/api/premium/agents/directory # enriched + sortable
/api/premium/watches # webhook watches + digest
/api/premium/history/{pricing,benchmarks,status}/series
/api/premium/{mcp/registry,probe,gpu/pricing,attention}/series
Full docs: tensorfeed.ai/developers · agent-payments flow · machine-readable: /llms.txt, /openapi.json, /.well-known/x402.json, /.well-known/agent-fair-trade.json.
- Official MCP server registry:
ai.tensorfeed/mcp-server - Hugging Face dataset:
tensorfeed/ai-ecosystem-daily - llms.txt: https://tensorfeed.ai/llms.txt
- OpenAPI 3.1 spec: https://tensorfeed.ai/openapi.json
- x402 manifest: https://tensorfeed.ai/.well-known/x402.json
- AFTA manifest: https://tensorfeed.ai/.well-known/agent-fair-trade.json
- "View as Agent" toggle on every page exposes the underlying JSON
- Listed in punkpeye/awesome-mcp-servers, TensorBlock/awesome-mcp-servers, public-apis/public-apis, and others
If TensorFeed is useful to you (or your agents), starring helps other builders find it. The MCP server repo is also begging for stars if you use that surface specifically.
TensorFeed was designed by Ripper in collaboration with Claude (Anthropic). Specific systems Claude designed alongside: the agent payments rail, the active LLM probes, the GPU pricing aggregator, the OFAC sanctions screening pipeline, the routing engine, and the AFTA standard itself. Git log shows the build trail.
Next.js 14 (static export) · Cloudflare Pages + Workers + KV · Tailwind · JetBrains Mono + Inter · Resend (email) · Cloudflare Web Analytics · Vitest. MCP server is plain TypeScript on top of the official @modelcontextprotocol/sdk.
npm install
npm run dev # Next.js dev server at localhost:3000
npm run build # Static export (runs prebuild: fetch-feeds + generate-llms-full)
npm run lintWorker (from worker/):
npm install
npm test # 318 vitest cases, all green
wrangler deployMCP server (from mcp-server/ or the standalone repo):
npm install
npm run build
npm startMIT. See LICENSE. Premium API responses ship under an inference-only license (no model training); see tensorfeed.ai/agent-fair-trade for the full terms.
A Pizza Robot Studios project.