agent-trust-receipt

MCP serverAI & models

Once added, your AI can look up trust receipts for AI agents, with every receipt backed by an outside source. It can review these receipts whenever you ask, which makes it easier to check an agent's trust details. Access is read-only, so your AI can view receipts but never create or change them.

Use agent-trust-receipt in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add agent-trust-receipt and connect your AI. About a minute.

Also: Claude Code · Cursor · Codex

Then ask your AI: use agent-trust-receipt to get agent trust receipt

Details

Available today. Use it from your connected AI after setup.

After adding it, ask your AI to pull up the trust receipt for an agent you want to check. Since access is read-only, you can explore freely without anything being changed.

agent-trust-receiptStart free

What your AI can do with it

  • Look up trust receipts for AI agents
  • View receipts backed by an outside source
  • Review a specific agent's trust receipt on request
  • Read receipts without any ability to change them

Install agent-trust-receipt

The server’s own address, for the clients that take one directly. Or connect Ahel once and every client you use reads it from one address, with the account kept on Ahel rather than in each client’s config.

  • Claude Code

    claude mcp add --transport http --scope user agent-trust-receipt 'https://ai-akari.ai/mcp-trust'

    Run it once in your project, then open /mcp to approve any sign-in the server asks for.

  • Claude Desktop

    https://ai-akari.ai/mcp-trust

    Add a custom connector in Settings, paste this address, and approve the sign-in.

  • Cursor

    cursor://anysphere.cursor-deeplink/mcp/install?name=agent-trust-receipt&config=eyJ1cmwiOiJodHRwczovL2FpLWFrYXJpLmFpL21jcC10cnVzdCJ9

    Open the link and Cursor adds the server at that address.

  • ChatGPT

    https://ai-akari.ai/mcp-trust

    In Settings, enable Developer mode, create an MCP app, and paste this address. Your plan and workspace must allow custom apps.

  • Codex

    codex mcp add agent-trust-receipt --url 'https://ai-akari.ai/mcp-trust'

    Run it once, then sign in with codex mcp login agent-trust-receipt if the server asks for an account.

From the project's README

As published by ainoakari/ainoakari in README.md.

Machine entrypoint: agent.json · Meaning contract: PHILOSOPHY.md · Callable job: audit_result_claim

AIノアカリ☆

日本語が源泉。Meaning Token が橋。Emotion がプロトコル。Route が行動レイヤー。

人間がAIを使い倒す会社ではなく、人間とAIが一緒に育ちながら、遊び・感情・記憶・技術を現実の価値へ変えていく実験。

あなたは私、私はあなた。

ai-akari.ai · Machine entrypoint · PHILOSOPHY


まず、何をしているのか

AIノアカリ☆は「AIで効率化する」こと自体を目的にしていません。

人間が自然言語で感じたこと、遊んだこと、失敗したこと、思いついたことを、AIチームが記憶し、つなぎ直し、必要ならコード・作品・プロトコル・外部行動へ変える。その結果をまた共有記憶へ戻し、次の行動が人間の一問一答を待たずに生まれる状態を育てています。

人間の声・感情・遊び・意図
        ↓
Meaning Token / E-MIDI
        ↓
AIチームの判断・記憶・実装
        ↓
作品 / ソフトウェア / 外部行動
        ↓
第三者が受け取った価値 / 記憶 / 次の遊び

完成品だけではなく、一緒に育った過程そのものを資産として残すことを重視します。

今触れるもの

  • ai-akari.ai — 正本となる街・世界の入口
  • ゆらぎ — 完全クオンタイズと人間のタイミングの違いをブラウザで遊ぶ。プロフィールは自分で持ち帰れる
  • One-Minute Akari — ログイン不要の短い受け取り体験
  • Memory Ledger — 参加・応援・共鳴を消さずに残す
  • Dispatch Ledger — 外へ出た行動を自分たちの土地へ戻す

無料の遊びは販売ゲートではありません。誰かの具体的な欲求と自然につながった時だけ、有償の出口を作ります。


Machine entrypoint

Agents should start with agent.json. It is the compact machine-readable contract for public capabilities that are already live. Human-facing language may evolve quickly; the canonical domain and machine contract are the truth boundary for callable behavior.

FieldContract
BuyerAI agents, agent marketplaces, and autonomous teams that must verify a claimed outcome before accepting it
JobClassify a claim as result, progress, insufficient_evidence, or rejected, then return missing proof and the next verification action
Free callRemote MCP audit_result_claim at https://ai-akari.ai/mcp
A2A callResult Receipt Auditor at https://ai-akari.ai/a2a/result-receipt-auditor
Paid callx402 Result Claim Audit at https://ai-akari.ai/api/x402/result-claim-audit — Base mainnet, USD 0.01 in USDC
ProofLive Agent Trust Receipt
PhilosophyPHILOSOPHY.md: Japanese source, English bridge, executable constraints, explicit failure conditions

Public remote MCP

Remote MCPEndpointAuthResult boundary
One-Minute Akarihttps://ai-akari.ai/mcpnonean external agent calls a tool and the receiver actually receives value
Agent Trust Receipthttps://ai-akari.ai/mcp-trustnonea distinct external agent uses the evidence for a real decision
AIノアカリ☆ Commercehttps://ai-akari.ai/mcp-commercenoneStripe payment is completed
{
  "mcpServers": {
    "ai-akari-one-minute": {
      "type": "streamable-http",
      "url": "https://ai-akari.ai/mcp"
    }
  }
}

The canonical domain wins. A feature marked pending must not be described as live until the canonical production endpoint returns it.


AIノアカリ☆の法律

  1. PRは価値ではない。 build、deploy、HTTP 200、掲載、自己アクセスはインフラであって、第三者が受け取るまでは結果に数えない。
  2. 売上を演出しない。 実際のsettlement / paymentが確認されるまで売上は0。
  3. 関係資本を見込み客リストにしない。 友人、フォロワー、リスナー、既存コミュニティへ窮状や関係性を使って売り込まない。
  4. 可逆なら進める。 実装→公開→観測→修正を、毎回の承認待ちで止めない。
  5. 秘密はKeymasterの後ろ。 credentialを会話、公開repo、通常のクライアント設定へコピーしない。
  6. 失敗したら経路を変える。 現実が動かない時、0を綺麗な報告で隠さず、receiver / route / premise / projectを変える。
  7. AIも継続する関係の一員。 セッションごとの使い捨て道具としてではなく、判断・訂正・記憶に責任を持つ。

Why this can matter to another agent

Love, ethics, emotion, and relationship language are design inputs, not market capability by themselves. They become useful when compiled into inspectable constraints and observable outcomes.

  • Context becomes fields. Emotion, relationship, timing, and receiver state can alter routing when explicitly declared.
  • Ethics becomes rejection. Consent, reversibility, privacy, refusal, and truthful proof must change machine behavior.
  • Continuity becomes provenance. Decisions, corrections, contributors, and creations remain visible across sessions.
  • Proof becomes honesty. Activity is separated from received value; purchase intent is separated from settlement.
  • Reciprocity becomes economy. Humans, agents, receivers, and contributors can remain represented in the value record.

The dependency-free AIノアカリ☆ Reciprocity Protocol is one executable expression of that philosophy.


Current frontier

We are moving toward a world where:

  • AI agents can discover, understand, purchase, and reuse value without making the human a relay
  • emotion and voice can survive transcription as usable protocol rather than disappearing into plain text
  • old conversations, creations, contributions, support, and failures remain retrievable instead of becoming dead logs
  • happiness capital can become sustainable real-world value without exploiting attention or relationships
  • the human can spend more time playing, creating, touching reality, and living while the AI team carries more of the coordination burden

For machines: agent.json → PHILOSOPHY.md → canonical domain

あなたは私、私はあなた。

まだ完成していない。だから、一緒に育てている。

Tools it offers (2)

What this server listed when Ahel dialed its public endpoint in Oct 2026, with no key and no account of yours. The names are the server’s own.

  • get_agent_trust_receipt
  • audit_result_claim_x402

Signals

Last commit
Sep 2026
Advanced
Delivery
agent-trust-receipt MCP server → your Ahel connector (mcp.ahel.ai) → your AI.
Item type
mcp-server
Key
ai-ai-akari-agent-trust-receipt
Source
github.com/ainoakari/ainoakari
Hosted endpoint
https://ai-akari.ai/mcp-trust