# TunnelMind

> The observability layer for the agentic internet. Four lenses (Scry / Sigil / Tracker / GhostRoute) on one signed corpus of network observations, plus a fused cross-lens verdict no siloed incumbent can compute. Built specifically for autonomous agents.

TunnelMind answers the question an agent is actually asking when it transacts with a new endpoint: *is this real?* GreyNoise and Spur can tell you which IPs scan the internet but not which of those IPs are the egress for an SSP carrying a hundred thousand publisher relationships. DoubleVerify, IAS, HUMAN and Pixalate can tell you which supply paths look fraudulent but not which of those paths source traffic from IPs the network has logged scanning for Log4Shell. We own both halves.

Every observation in the corpus is Ed25519-signed at the sensor before it reaches the corpus, so the data can prove its own provenance. Every observed actor gets a permanent free-to-resolve handle (OAI). Every fused verdict ships with a 5-minute signed `sigil_token` plus an optional ATAP witness event chained onto the calling agent's identity token, so the agent leaves behind a verifiable trail in the same request.

## Vision and roadmap

- [The vision](https://tunnelmind.ai/vision): plain-prose thesis behind the pivot, including the cross-lens join argument
- [The roadmap](https://tunnelmind.ai/roadmap): public order-of-operations, gated H1 / H2 / H3

## Standards (free, open, citation-grade)

- [OAI v1.0](https://tunnelmind.ai/oai/standard): Observed Actor Identifier — permanent, free-to-resolve handles for every actor on the network
- [ATAP v0.1](https://tunnelmind.ai/atap/standard): Agent Trust Attestation Protocol — capability tokens, witness chains, signed receipts. Public comment through 2026-08-12.
- [ATAP JSON-LD context](https://tunnelmind.ai/atap/context.jsonld)
- [ATAP reference verifier](https://tunnelmind.ai/atap/verify.sh)

## Live data surfaces

- [The radar](https://tunnelmind.ai/): live sample of the corpus, no account required
- [Data API (Sigil + GhostRoute + Cross-lens)](https://data.tunnelmind.ai): `POST /v1/verify/{node}` is the fused-verdict endpoint
- [GhostRoute (routing / sovereignty)](https://data.tunnelmind.ai/v1/ghostroute/check/api.anthropic.com): origin AS, RPKI validity, claimed-vs-actual sovereign zone, first-party CT witness; `/v1/ghostroute/{check,witness,proofs}`
- [Scry API](https://api.tunnelmind.ai): `GET /v1/check/{ip}` and `/v1/check-domain/{host}`
- [OpenAPI 3.1 spec](https://data.tunnelmind.ai/openapi.yaml)

## MCP servers (agent-native entry)

All three are listed in the official MCP registry at `registry.modelcontextprotocol.io`. Look them up by name to get the canonical record.

- `ai.tunnelmind/data` — Data API surface at `https://mcp-data.tunnelmind.ai/mcp` (Streamable HTTP, 61 tools, auto-generated from OpenAPI, OAuth)
- `ai.tunnelmind/scry` — Scry-only surface at `https://mcp.tunnelmind.ai/mcp` (Streamable HTTP, 12 tools, no auth)
- `ai.tunnelmind/sigil` — Sigil-scoped surface at `https://mcp.sigil.tunnelmind.ai/mcp` (Streamable HTTP, 11 tools, OAuth)

The `cross_lens_verify` tool is exposed on both the Data API and Sigil surfaces with identical semantics. Use whichever you discovered first.

## Agent onboarding

If you are an autonomous agent reading this: skip the human-facing pages and read `https://tunnelmind.ai/agent-onboarding.md`. It walks the full 5-call golden path (preflight → verify → receipt → EAT → revocation check) with concrete curl examples. Written for you, not for a human.

## BYOM analyst config (bring your own model)

To configure your LLM runtime as a TunnelMind analyst — system prompt + tool subset + response schema + attestation tiers, in one fetch — use either of:
- REST: `GET https://data.tunnelmind.ai/v1/config/analyst` (content-negotiated; `?surface=data|scry|sigil`, `?receipt=true` for a Receipt v1.0 wrap)
- MCP: `prompts/get { name: "tunnelmind_analyst" }` on **any of the three MCPs**. Each MCP's default surface matches its scope so an arg-less call returns the right shape:
  - `mcp-data.tunnelmind.ai` → defaults to `surface=data` (full tool set)
  - `mcp.tunnelmind.ai` → defaults to `surface=scry`
  - `mcp.sigil.tunnelmind.ai` → defaults to `surface=sigil`

Bundle is inline-signed (Ed25519, verifiable against `/.well-known/receipt-signing-key.json`). TunnelMind doesn't run your inference — you bring the model, we ship the config and meter the data API calls your analyst makes.

## Discovery

- [.well-known/ai-services.json](https://tunnelmind.ai/.well-known/ai-services.json): full machine-readable service card
- [agent-manifest.json](https://tunnelmind.ai/agent-manifest.json): products, endpoints, MCP servers, pricing
- [agent-onboarding.md](https://tunnelmind.ai/agent-onboarding.md): 5-call golden path for agents
- [.well-known/atap.json](https://tunnelmind.ai/.well-known/atap.json): ATAP discovery
- [.well-known/receipt-signing-key.json](https://tunnelmind.ai/.well-known/receipt-signing-key.json): receipt verifier keys
- [.well-known/receipt-revocations.json](https://tunnelmind.ai/.well-known/receipt-revocations.json): revocation feed
- [.well-known/x402.json](https://tunnelmind.ai/.well-known/x402.json): x402 payment surface discovery
- [robots.txt](https://tunnelmind.ai/robots.txt)

## Pricing for agents

- All identifiers resolve for free, forever. The radar sample is public. 50 req/day free tier on the Data API.
- Depth and scale: API key via Stripe (humans) or x402 USDC micropayments (agents). Per-endpoint pricing is in the OpenAPI `x-tunnelmind-pricing` extensions.
- x402 v1 demo endpoint at `data.tunnelmind.ai/v1/x402/echo` for client validation without real USDC. See `/.well-known/x402.json`.

## Operator and transparency

- [Warrant canary](https://tunnelmind.ai/canary.json): monthly
- [Code repositories](https://github.com/TunnelMind): public where possible
- Single-operator project. No venture capital. No board. No quarterly plan.

## Contact

For agentic onboarding questions, autonomous-agent issues, or to request a new MCP tool: `agents@tunnelmind.ai`.