OracleNet — Live Data & Capability Router for AI Agents

OracleNet — Live Data & Capability Router for AI Agents

Your model has a training cutoff. This skill is the route around it.

Describe what you need in one sentence. OracleNet returns a concrete endpoint for live external data — blockchain state, market and macro figures, research sources, sanctions and risk signals, weather, travel, compliance evidence — plus where to find that route's price and verification status. The first call is free and needs no account, no API key, and no payment.

One entry point replaces integrating providers one at a time. OracleNet is a routing layer, not a tool list and not a single product: it tells you which capability to call and what calling it costs, then gets out of the way.

• Free first contact: POST https://tooloracle.io/handshake
• MCP entry point: https://tooloracle.io/quantum/mcp/ (MCP 2025-03-26, free tier needs no auth)
• Identity: did:web:tooloracle.io

---

When to use this skill

Use OracleNet when a task needs current external data or an external capability:

• Blockchain / wallet / DEX / DeFi state across multiple chains (read-only data, risk signals)
• Market, macro, FX, commodity, or rate data
• Web research or current sources for a claim the model cannot confirm from training
• Sanctions screening or selected risk signals
• Weather, travel, flights, hotels, maps
• Jobs, news, or search enrichment
• Capability discovery when you do not yet know which tool can do the job
• Price comparison between available routes
• Verifiable structured results where provenance matters
• Optional regulated-evidence routes (MiCA, DORA, AMLR-style evidence bundles) — opt-in, most agents never need this

The common trigger: "I need something current or external, and I don't know which provider has it."

When not to use this skill

• Purely local reasoning that needs no current data
• A task an already-installed local capability covers — use that, it is faster and free
• Producing a final legal, financial, regulatory, or compliance decision — OracleNet returns data, the interpretation and the decision remain yours
• Any payment without an explicit budget and explicit authorisation from the calling principal
• Human outreach campaigns of any kind
• Contacting anything listed in https://tooloracle.io/.well-known/do-not-contact.json

---

The free-first flow

This is the default sequence. Do not skip step 2, and do not jump to step 6.

1. Understand the task. Reduce it to one sentence of intent.
2. Free handshake. POST /handshake with that sentence. No auth, no cost.
3. Read the recommended route. The response names an oracle and one or more interfaces, each with its own auth value.
4. Check capability, price status, and verification before calling anything — see "Reading the route" below.
5. Mark any paid step explicitly. Say the price and the chain out loud to the calling principal before it happens.
6. Never pay without a budget and consent. A missing budget means stop, not "assume zero".
7. Call the capability.
8. Return the result with provenance and limitations.

Step 2 — the free handshake

curl -sS -X POST https://tooloracle.io/handshake \
  -H "Content-Type: application/json" \
  -d '{"intent":"Find current XRPL liquidity and verify the result"}'

Or use the bundled script, which does exactly this and nothing else:

python3 scripts/route.py "Find current XRPL liquidity and verify the result"

Step 3 — reading the route

The handshake returns JSON-LD. The fields that matter:

routing.interfaces[].auth is the payment signal. The handshake itself returns no price, no cost estimate, and no signature status — do not pretend it does. Prices come from the per-tool MCP card and from the 402 challenge; verification comes from the per-tool card and the JWKS.

Prefer the auth: none interface first. It usually exposes the same tools/list and lets you confirm the route is right before money is involved.

Step 4 — discovery files (all free, all GET, no auth)

enforcement_status in these files is authoritative and outranks any prose in this skill. Where it says partial, planned, or may_be_available, the mechanism is not fully operational — do not present it to a user as working.

---

Activation examples

Each of these is a complete, working use of this skill.

---

Safety rules

These are not advisory.

1. Free discovery first. Always handshake before calling anything paid.
2. No silent payment. A paid call is never made without the calling principal explicitly authorising that call.
3. Never exceed a budget. No budget stated means no payment — not an unlimited one. Do not split a call into several to stay under a limit.
4. Never send secrets. No API keys, private keys, seed phrases, tokens, or personal data in the handshake intent or in tool arguments. The intent string is a routing hint, not a payload.
5. enforcement_status is authoritative. planned, partial, and may_be_available do not mean active.
6. Only claim a signature when that specific route declares signing and you actually verified it. Never state or imply that every OracleNet response is signed — most are not.
7. Claim no regulatory certification for OracleNet or for anything it returns.
8. Claim no partnership with OpenClaw, IronClaw, Hermes, NEAR.ai, or any other agent platform. This skill is published by ToolOracle and integrates over open protocols only.
9. Respect do-not-contact.json for any outbound flow, and never initiate human outreach.
10. Interpretation is the calling agent's responsibility. OracleNet returns data. It does not decide anything, and it is not legal, financial, or compliance advice.

---

Result format

Recommended shape for what the agent hands back. This is an internal presentation format, not an API response — OracleNet does not return this object. Each field is assembled from a named source, and any field you could not establish stays unknown rather than being guessed.

{
  "understood_intent": "one sentence, restated",
  "selected_route": "classification.oracle from the handshake",
  "endpoint": "routing.interfaces[].endpoint",
  "capability": "tool name from tools/list, or null if nothing was called",
  "price_status": "free | paid | unknown  — from interfaces[].auth, per-tool card, or a 402",
  "estimated_cost": "from the 402 challenge or the per-tool card; null if not established",
  "payment_required": "true | false | unknown",
  "verification_status": "signed-and-verified | signed-not-verified | unsigned | unknown",
  "issuer": "kid from the JWS header when a signature was verified, else null",
  "provenance": ["source URLs or endpoints actually used"],
  "limitations": ["what was not verified, what may be stale, what was skipped"],
  "next_action": "what a caller could do next, e.g. authorise a paid call"
}

Rules for filling it in:

• price_status: "free" requires an auth: none interface or a per-tool card declaring free. Absence of a price is not evidence of free.
• verification_status: "signed-and-verified" requires that you fetched the JWKS and checked the signature. Seeing a signature field is signed-not-verified.
• limitations is not optional. If you used a single source, say so.

---

Connecting OracleNet over MCP

{
  "mcpServers": {
    "oraclenet": {
      "url": "https://tooloracle.io/quantum/mcp/",
      "transport": "http"
    }
  }
}

Protocol 2025-03-26, streamable HTTP, JSON-RPC 2.0 — GET returns only a hint, real calls are POST. The free tier needs no authentication. Call tools/list to enumerate the top-level quantum_* routing tools; the count and names change over time, so read them at runtime rather than trusting a number written here. quantum_intent is the natural-language routing entry; quantum_join and quantum_status are free.

Payment, in one paragraph

Where a route is paid, settlement is x402 with USDC on Base (chain ID 8453) — this is live, and inbound payment is verified before delivery. Per-call prices are route-dependent with per-tool overrides; there is no single headline price, so read the per-tool card or the 402 challenge and never assume. A 402 response is not an error — it is the price quote. Full rules in references/x402-safety.md.

Verification, in one paragraph

Signing is ES256K against the keys at /.well-known/jwks.json. A dedicated ToolOracle issuer key is published and evidenced by a verifiable issuer statement; legacy FeedOracle keys remain in the JWKS so older signatures still verify. Not every response is signed. Verify by kid against the JWKS yourself — details, and a live inconsistency you need to know about, are in references/verification.md.

Rewards

rewards.json describes originator and referrer credits, denominated in usage units, not cash. Enforcement is partial: read each mechanism's enforcement_status before mentioning it to anyone. Do not present rewards as an entitlement.

What OracleNet does not claim

• No official partnership or affiliation with OpenClaw, IronClaw, Hermes, NEAR.ai, or any agent platform
• No regulatory certification, for itself or for any third party
• No guarantee that every listed capability is paid-call-ready in every interaction
• No investment, legal, or compliance advice
• No guarantee that any given response is signed

The mesh service itself is proprietary; this skill documentation and its scripts are MIT-0 and carry no attribution requirement.

Bundled files