openzoo
Local x402-paying proxy and MCP server for openzoo.fun, enabling any OpenAI-compatible tool to pay for AI calls using a local burner wallet. It provides tools like zoo_ask for large-corpus questions, model listing, and wallet management.
README
openzoo
Local x402-paying proxy + MCP server for openzoo.fun — point any OpenAI-compatible harness (Cursor, Claude Code, aider, raw SDKs) at localhost and it pays the zoo per call from a local burner wallet. No account, no API key, no crypto knowledge required by your tools.
npx openzoo
That's it. First run generates a burner wallet at ~/.openzoo/wallet.json (chmod 600), prints the funding address, and starts the proxy:
base_url = http://localhost:8402/v1
api_key = sk-openzoo # any value works; the zoo takes payment, not keys
Every request forwards to the zoo. When the zoo answers 402 Payment Required, the proxy reads the quote, signs a token transfer from your local wallet, retries with the X-PAYMENT header, and streams the response back — SSE included, unbuffered. Each payment prints a one-line receipt:
paid $0.002137 (9.5× cheaper than direct) · rail solana · tx 5Kd…
60-second quickstarts
Cursor (Settings → Models → OpenAI API): set Override OpenAI Base URL to http://localhost:8402/v1, API key sk-openzoo. (Cursor Hobby can't BYOK; Pro can.)
Claude Code / any OpenAI-env tool:
export OPENAI_BASE_URL=http://localhost:8402/v1
export OPENAI_API_KEY=sk-openzoo
aider:
aider --openai-api-base http://localhost:8402/v1 --openai-api-key sk-openzoo --model openai/nvidia/nemotron-3.5-lightning
curl:
curl http://localhost:8402/v1/chat/completions -H 'Content-Type: application/json' \
-d '{"model":"nvidia/nemotron-3.5-lightning","messages":[{"role":"user","content":"hi"}]}'
OpenAI SDK (Python):
from openai import OpenAI
client = OpenAI(base_url="http://localhost:8402/v1", api_key="sk-openzoo")
r = client.chat.completions.create(model="nvidia/nemotron-3.5-lightning",
messages=[{"role": "user", "content": "hi"}])
OpenAI SDK (JS/TS):
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "http://localhost:8402/v1", apiKey: "sk-openzoo" });
const r = await client.chat.completions.create({
model: "nvidia/nemotron-3.5-lightning",
messages: [{ role: "user", content: "hi" }],
});
Model ids come from GET http://localhost:8402/v1/models (free, no payment). Requires Node ≥ 18 (for npx openzoo itself; your harness can be anything).
Use as MCP
The same package is an MCP server sharing the same wallet and payment core:
claude mcp add openzoo -- npx -y openzoo mcp
Claude Desktop — add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json, Windows: %APPDATA%\Claude\claude_desktop_config.json), then restart the app:
{ "mcpServers": { "openzoo": { "command": "npx", "args": ["-y", "openzoo", "mcp"] } } }
Cursor — Settings → MCP → Add new global MCP server, or drop the same JSON into ~/.cursor/mcp.json (per-project: .cursor/mcp.json).
Windsurf / Cline / any MCP host — same shape: command npx, args ["-y", "openzoo", "mcp"], stdio transport.
Tools:
zoo_ask—{prompt, corpus?, model?, max_tokens?}. The flagship: hand it a hugecorpus(hundreds of thousands to ~1M tokens — a body the model itself would refuse) and a question; the zoo's leCore memory spills the corpus so the model reads only a few thousand tokens. Returns the answer plus the receipt (billedUsd,savesVsDirect, tokens actually read). An MCP-capable agent can delegate a giant-context question without touching its own model config.zoo_models— list the zoo's models and pricing (free).zoo_wallet— funding address, balances, and this session's receipts.
The demo
npx openzoo demo
Builds a ~965k-token document with one planted fact, shows that buying direct refuses it ("The input token count exceeds the maximum number of tokens allowed." — recorded, see benches.openzoo.fun), binds the body ONCE to the zoo's holographic memory, then pays for a question that ships alone:
bound once in 14.8s: 3.7MB → context ctx_01KZZY8YQE…
quote for the ask: $0.000480 · pricing=markup (3× a tiny body — the 3.7MB corpus is not re-priced)
If the wallet is funded with USDC or TOKEN it pays (capped at OPENZOO_DEMO_MAX_USD, default $0.01) and prints the answer, the tokens the model actually read, and the receipt. If not, it prints exactly what to fund. Long waits (the one-time upload, pricing, payment, the answer) show a live progress line with stage + elapsed seconds.
Run it twice. The second run finds the corpus in the local manifest and never uploads it:
★ corpus already bound (context ctx_01KZZY8YQE…) — skipped the 3.7MB upload entirely.
...
★ corpus already bound — skipped 3.7MB upload; whole run took 4.3s and cost $0.000480
Repeat calls are near-free (0.3.0: the body never ships twice)
The zoo keeps your corpus in leCore holographic memory; the shim keeps a manifest at ~/.openzoo/contexts.json (chmod 600) mapping sha256(corpus) → the zoo's context_id, scoped per API base. When a request carries a corpus the manifest already knows:
- nothing big is uploaded — the ask ships alone with an
X-HRR-Contextheader, - the 402 prices the tiny ask (markup basis, honestly labeled
pricing=markup), typically a few hundredths of a cent instead of re-pricing megabytes, - the answer still comes from your corpus — the zoo recalls the relevant slices server-side.
This works in all three fronts: the proxy (a big pasted body in the last message is split at its last blank line, bound once, and reused on every later call — even with a different question), the MCP zoo_ask corpus parameter, and the demo. If the zoo ever forgets a context (sidecar wipe), the gateway answers 404 before any payment and the shim transparently re-binds once and retries — a stale manifest never fails a call.
Manage it:
npx openzoo contexts # list bound corpora (hash, context id, age, api base)
npx openzoo contexts --forget 60464d # drop one by hash prefix
npx openzoo contexts --forget all # drop everything
Opt out with OPENZOO_NO_CONTEXT_CACHE=1 (always ship the full body); tune the threshold with OPENZOO_CONTEXT_MIN_CHARS (default 16384 chars).
Using openzoo from a cloud IDE (Cursor, Windsurf, hosted agents)
Some IDEs run their model calls from their own servers, not your machine. Point one of those at http://localhost:8402 and you get, verbatim:
Provider returned error: Access to private networks is forbidden
That is reachability, not payment — their cloud cannot dial your laptop. Give it a public URL instead:
npx openzoo tunnel
It installs cloudflared itself (one-time, cached in ~/.openzoo/bin; a Cloudflare account is not needed), opens a quick tunnel to your local proxy, and prints:
base_url = https://<random>.trycloudflare.com/v1
api_key = oz_<random> # in tunnel mode the key is REAL auth
Tunnel mode is the one mode where the api key matters. A public URL in front of a wallet is a public URL in front of your money, so:
- every request without
Authorization: Bearer <that key>is refused 401 before anything is forwarded, quoted or paid; - the per-call cap (
OPENZOO_MAX_USD_PER_CALL, default $0.50) still applies; - a session ceiling stops all spending at
OPENZOO_TUNNEL_MAX_USD(default $1.00); - every served request prints its receipt and the running session total;
- the URL dies when you Ctrl-C, and the session's total spend is printed on exit.
Pin the key with OPENZOO_TUNNEL_TOKEN if your IDE stores it. Keys never leave your machine either way — the tunnel forwards to the same local proxy, which signs with the same local wallet.
MCP clients don't need any of this. If your tool speaks MCP, use the hosted server at https://mcp.openzoo.fun/mcp — cloud-reachable, no local process, mint a wallet with zoo_wallet.
The wallet model
-
Burner, local, yours. A keypair in
~/.openzoo/wallet.json, created on first run, chmod 600. Keys never leave your machine — the zoo only ever sees signed transfers. -
Fund it with USDC or TOKEN. Send a few cents of either to the address
npx openzoo addressprints:- USDC —
EPjFWdd5AufqSSqeM2qN1xzybapC8G4wEGGkZwyTDt1v - TOKEN —
EVULoNF4DeMBN4dGiZiDfpiiTfNZgoCvXWWgaV3epump
Both are first-class: each is the underlying of a live Solana rail the zoo quotes (USDC → yUSDCx, TOKEN → wTOKENx). That's the whole funding story — the shim converts whichever the 402 quotes internally, at payment time, for exactly the amount needed.
- USDC —
-
SOL is optional but nice. The gateway sponsors payment-transaction fees. If the wallet holds a pinch of SOL (~0.003), the internal conversion settles as its own transaction first; with zero SOL the conversion rides inside the gateway-sponsored payment transaction instead.
-
The same wallet file also holds an EVM key.
npx openzoo addressprints both addresses: the Solana one, and the EVM one used by the Base and Robinhood Chain rails. Fund Base with plain USDC on Base (nothing is converted); for Robinhood Chain, hold the plain token you want to pay with (USDG or a quoted memecoin) plus a sliver of RH ETH for gas — conversion happens automatically at payment time. -
Spend caps. The proxy refuses any single quote above
OPENZOO_MAX_USD_PER_CALL(default $0.50). -
npx openzoo balance— funds on every rail, grouped per chain: Solana (USDC + TOKEN + SOL), Base (USDC + ETH), Robinhood Chain (USDG, the ODDBALLER / IOU / ROBINHOODS memecoins, ETH). USD value where a price is known;$?where not.npx openzoo address— both funding addresses.
Honest pricing note
Two bases, reported per call in the 402 (extra.pricing):
- Short prompts price at a markup (3× provider cost) — there's nothing to spill, you're paying for passthrough.
- Big bodies price at a counterfactual discount (~10× cheaper than buying the same call direct) — the zoo's leCore memory means it never forwards your whole body upstream, and passes the savings on. Measured numbers at benches.openzoo.fun.
- Asks against a bound corpus price on the markup basis (3× a tiny body — the receipt says
pricing=markup). That is not a discount trick: 3× of a few hundred tokens is normally far below even the counterfactual price of shipping the corpus, which is the whole point of binding once.
The receipt names which base you got; extra.directUsd / extra.savesVsDirect let you check the math.
Payment rails
All three rails have settled real payments (2026-08-14):
| rail | network | status |
|---|---|---|
| Solana (default) | solana:5eykt… |
live — Token-2022 TransferChecked, partial-signed, gateway pays fees. Settles daily; tested end-to-end against the production 402. Settlement uses a wrapped settlement mint as internal plumbing; you only ever hold and send USDC or TOKEN. |
| Base | eip155:8453 |
live — standard x402 EIP-3009 transferWithAuthorization against native USDC, batched settle through the facilitator. Fund the wallet's EVM address with USDC on Base; nothing is converted. |
| Robinhood Chain | eip155:4663 |
live — EIP-3009, batched settle through the facilitator. Hold the plain token a row is quoted in (USDG, or the ODDBALLER / IOU / ROBINHOODS memecoins) and the shim converts exactly enough at payment time, automatically — two small on-chain steps paid from the wallet's own RH ETH. No gas? The error says exactly how much ETH to send and where. Default rail selection skips this chain unless OPENZOO_ENABLE_RH=1; OPENZOO_RAIL=robinhood forces it outright. |
npx openzoo prints the rails off a live 402 at startup, and the funding line is derived from those rails — so a new chain (the wrapped RH memecoin twins are in-deploy, for example) shows up without this package shipping again.
The rail is chosen from the 402's accepts[] itself (Solana first). Steer it with OPENZOO_RAIL=solana|base|robinhood — the picker then uses only that rail, and errors clearly if the live 402 does not offer it. Amounts are always taken as raw units from the 402, and Solana decimals are read from the mint on-chain — never hardcoded. (The zoo's own pasted prompt used to hardcode decimals = 6; this package deliberately does not copy the bug.)
Configuration
| env | default | |
|---|---|---|
OPENZOO_PORT |
8402 |
proxy port |
OPENZOO_API_BASE |
https://x402-tokens.fly.dev |
the zoo's door |
OPENZOO_RPC |
mainnet-beta public RPC | Solana RPC |
OPENZOO_TOKEN |
(internal) | preferred 402 rail — leave unset |
OPENZOO_RAIL |
(unset) | force a rail: solana | base | robinhood. Errors if the live 402 doesn't offer it |
OPENZOO_BASE_RPC |
https://mainnet.base.org |
Base RPC (balances / preflight) |
OPENZOO_RH_RPC |
https://rpc.mainnet.chain.robinhood.com |
Robinhood Chain RPC (balances / preflight) |
OPENZOO_WALLET |
~/.openzoo/wallet.json |
wallet path |
OPENZOO_MAX_USD_PER_CALL |
0.5 |
refuse quotes above this |
OPENZOO_DEMO_MAX_USD |
0.01 |
demo spend cap |
OPENZOO_CONTEXT_MIN_CHARS |
16384 |
bodies bigger than this bind once + reuse |
OPENZOO_NO_CONTEXT_CACHE |
0 |
set 1 to always ship the full body |
OPENZOO_ENABLE_RH |
0 |
let default selection fall through to the Robinhood rail (OPENZOO_RAIL=robinhood forces it without this) |
What's tested
- Unit tests run against a captured production 402 body (
test/fixtures/live-402.json): parsing, rail selection, and byte-level verification of the signed Solana transaction and the EVM typed-data payload. - Live-verified:
/v1/modelsthrough the proxy (200), 402 parse + quote against mainnet, on-chain mint/decimals read, the internal USDC conversion (funded from plain USDC, settled on mainnet), and a real settled paid call — the full demo ran end-to-end against the production gateway: quote, automatic top-up from USDC, payment accepted by the facilitator, answer + receipt returned.
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.