XAIP — Trust Infrastructure for AI Agents

Not another agent framework. A portable signed receipt layer underneath them.

XAIP is a provider-neutral signed execution evidence layer for AI agent tool calls. It records signed receipts across MCP, LangChain.js, OpenAI-compatible tool-call loops, and other runtimes, then exposes historical execution evidence that agents, developers, or policy layers can inspect before delegation.

Receipts are the primary artifact. Trust scores are one derived view over those receipts — not a claim of absolute safety or correctness.

Live demo: three candidate servers, side-by-side comparison without and with XAIP. Open in browser →

Provider-agnostic by design. XAIP is a trust layer for any tool-using agent. The reference implementation and live data start with MCP (Model Context Protocol) — because that's where the broadest fleet of public tool servers exists today — but the receipt format, signing, and scoring apply equally to LangChain tools, OpenAI function calling, A2A, and proprietary agent stacks. MCP is the first integration, not the only one.

Live dashboard: https://xkumakichi.github.io/xaip-protocol/ — current public trust scores, auto-refreshed, no auth. The current public dataset is MCP-heavy because MCP was the first integration target.

Entry points

• Trust Evidence Before Delegation — one-screen demo, live POST /v1/select against three contrasting candidates.
• Browser playground — read-only demo of trust-aware selection.
• 60-second overview — the problem XAIP is trying to address.
• Future direction — long-term hypothesis, open questions, and current research asks.
• Agent Trust Check design — planned diagnostic concept.
• Class-aware scoring design — future design note; not live scoring behavior.
• Emit receipts from anything — how to produce XAIP receipts from any tool system.
• Run xaip-caller — contribute signed receipts without running MCP.

Try It Now

The API is live. No signup, no API key.

# Check trust score for a scored tool server
curl https://xaip-trust-api.kuma-github.workers.dev/v1/trust/context7

# Batch query
curl "https://xaip-trust-api.kuma-github.workers.dev/v1/trust?slugs=context7,sequential-thinking,filesystem"

# Decision engine: rank candidates by available execution evidence
curl -X POST https://xaip-trust-api.kuma-github.workers.dev/v1/select \
  -H "Content-Type: application/json" \
  -d '{"task":"Fetch React docs","candidates":["context7","sequential-thinking","unknown-server"]}'

The /v1/select response tells you which server to use, why, and what would happen without XAIP:

{
  "selected": "context7",
  "reason": "Highest trust among scored candidates based on current verified receipts",
  "rejected": [{ "slug": "unknown-server", "reason": "unscored — no execution evidence available" }],
  "withoutXAIP": "Random selection would pick an unscored server 33% of the time — no execution evidence available"
}

The Problem

Without trust scores, your agent is gambling:

┌────────────────┬────────────────┬───────────┬──────────────┐
│ Strategy       │ Server Hit     │ Success   │ Latency      │
├────────────────┼────────────────┼───────────┼──────────────┤
│ With XAIP      │ context7       │ ✓         │ ~3s          │
│ Random         │ unknown-mcp    │ ✗ error   │ ~8s (wasted) │
│ Try all (seq)  │ 3 servers      │ 1/3       │ ~11s total   │
└────────────────┴────────────────┴───────────┴──────────────┘

XAIP helps agents prefer candidates with stronger available execution evidence, skip unscored candidates when appropriate, and reduce avoidable failed calls.

How It Works

1. Select    POST /v1/select → ranks candidates by available execution evidence
2. Execute   Your agent calls the selected tool server
3. Report    POST /receipts → signed execution receipt feeds back into trust scores

Every execution receipt is Ed25519-signed and verified. Trust scores are computed using a Bayesian model with caller diversity weighting — not self-reported metrics.

Quick Start

Run the end-to-end demo

git clone https://github.com/xkumakichi/xaip-protocol.git
cd xaip-protocol/demo
npm install
npx tsx dogfood.ts

This demo:

1. Asks XAIP to rank candidate servers for "Fetch React hooks documentation" by available execution evidence
2. Connects to the selected MCP server and executes real tool calls
3. Submits a signed execution receipt to the Aggregator
4. Shows the updated trust score

Decision quality demo

Compare blind selection strategies against XAIP-guided selection using a static trust snapshot and fixed candidate sets:

cd demo
npm run blind-vs-xaip

This is a deterministic local replay. It does not perform live tool execution, post receipts, or call any external API. See docs/blind-vs-xaip-demo.md for scope, metrics, and limitations.

In the included snapshot replay:

Strategy	Risky pick rate	Eligible pick rate
Random	71.4%	28.6%
Fixed-order	85.7%	14.3%
XAIP	14.3%	85.7%

risky_pick = selected candidate was low_trust or unscored in the snapshot. fixed-order models an agent that accepts the upstream planner's candidate order without runtime trust data. The claim is limited to this fixed candidate set and static trust snapshot — not a guarantee of real-world execution improvement.

Become an independent caller

Want the trust graph to depend on more than one operator? Run a caller yourself. No account, no approval, no API key — the aggregator verifies signatures from any valid keypair.

Fastest — zero-install, 30 seconds:

npx xaip-caller

Signs receipts for a handful of real HTTP tool calls and POSTs them. Demonstrates that XAIP works beyond MCP — any HTTP tool can participate. See clients/caller. See Run xaip-caller for Windows notes and external receipt contribution details.

Full path — MCP servers, 5 minutes:

Clone the repo and run the auto-collector against real MCP servers. Your caller DID contributes to the diversity of every scored MCP tool. See docs/contributor/run-a-caller.md.

Use the SDK

npm install xaip-sdk

import { XAIPClient } from "xaip-sdk";

const client = new XAIPClient();

// Rank candidates by available execution evidence
const decision = await client.select({
  task: "Fetch React documentation",
  candidates: ["context7", "sequential-thinking", "unknown-server"],
});

console.log(decision.selected);    // "context7"
console.log(decision.withoutXAIP); // "Random selection would pick an unscored server 33% of the time..."

MCP Server

Use XAIP directly from Claude, Cursor, or any MCP-compatible AI agent:

npx xaip-mcp-trust

4 tools: xaip_list_servers, xaip_check_trust, xaip_select, xaip_report

Add to Claude Code (~/.claude/claude_desktop_config.json):

{
  "mcpServers": {
    "xaip-trust": {
      "command": "npx",
      "args": ["-y", "xaip-mcp-trust"]
    }
  }
}

npm: xaip-mcp-trust

API Reference

Method	Endpoint	Description
GET	/v1/servers	List all scored servers with trust data
GET	/v1/trust/:slug	Trust score for a single scored server
GET	/v1/trust?slugs=a,b,c	Batch trust scores (max 50)
POST	/v1/select	Decision engine — rank candidates by available execution evidence
GET	/health	Liveness probe

Base URL: https://xaip-trust-api.kuma-github.workers.dev

Trust Score Response

Field	Type	Description
trust	number | null	0.0–1.0 score, null if unscored
verdict	string	trusted ≥0.7 · caution 0.4–0.7 · low_trust <0.4 · unscored
receipts	number	Total verified execution receipts
confidence	number | null	Statistical confidence: min(1, receipts/100)
riskFlags	string[]	Detected risk indicators
computedFrom	string	Data provenance description

Decision Engine (POST /v1/select)

Request:

{
  "task": "description of what your agent needs to do",
  "candidates": ["server-a", "server-b", "server-c"],
  "mode": "relative"
}

• mode: "relative" (default) — always selects the best available, even if below threshold
• mode: "strict" — rejects all candidates below caution threshold

Architecture

┌──────────────────────────────────────────────────────────┐
│  Your AI Agent                                           │
│  ┌──────────┐   ┌───────────┐   ┌─────────────────────┐ │
│  │ Select   │──▶│ Execute   │──▶│ Report Receipt      │ │
│  │ (Trust   │   │ (MCP call)│   │ (Ed25519 signed)    │ │
│  │  API)    │   └───────────┘   └──────────┬──────────┘ │
│  └────┬─────┘                              │            │
└───────┼────────────────────────────────────┼────────────┘
        │                                    │
        ▼                                    ▼
┌───────────────┐                 ┌──────────────────────┐
│  Trust API    │◀────────────────│  Aggregator (BFT)    │
│  + Decision   │  Service        │  Cloudflare D1       │
│    Engine     │  Binding        │  Ed25519 verification│
└───────────────┘                 │  Bayesian scoring    │
                                  └──────────────────────┘

Trust Model:

• Bayesian Beta distribution (prior varies by DID method)
• Caller diversity weighting (prevents single-caller gaming)
• Co-signature factor (dual Ed25519: agent + caller)
• BFT-capable federation with MAD outlier detection across aggregator nodes; the current public deployment is a single aggregator (quorum:1)

Infrastructure:

• Cloudflare Workers (global edge, <50ms latency)
• Cloudflare D1 (SQLite at edge) for receipt storage
• Service Bindings for Worker-to-Worker communication

Optional Ledger-Backed Identity

XAIP is not a blockchain protocol and not a payment rail.

The current identity model supports multiple DID methods, including did:key, did:web, and ledger-backed identifiers such as did:xrpl. Trust scores are derived from signed execution receipts — not token holdings, payments, or chain affiliation.

The default priors below are deployment policy, not a universal claim about trust:

DID Method	Default Prior	Use Case
did:xrpl	[5, 1]	Ledger-backed agents (one externally anchored option)
did:web	[2, 1]	Domain-verified servers
did:key	[1, 1]	Anonymous / new agents

Ledger-backed identities may be useful when an agent needs an externally anchored identity. XAIP itself does not require any ledger.

Data

Trust scores are computed from real execution data, not synthetic benchmarks:

• ~3,100 signed receipts across 10 scored MCP servers as of 2026-05-08: context7, sequential-thinking, memory, filesystem, everything, fetch, sqlite, git, puppeteer, playwright
• Automated daily data collection via GitHub Actions
• Scores update with every new execution receipt; see the live dashboard/API for current values

# See all scored servers
curl https://xaip-trust-api.kuma-github.workers.dev/v1/servers

Works With

Runtime / integration	Status	How
MCP (Model Context Protocol)	Live public dataset (10 servers, ~3,100 signed receipts as of 2026-05-08)	xaip-claude-hook, xaip-sdk, xaip-mcp-trust
LangChain.js	Tested preview + live receipts landed	xaip-langchain callback handler
OpenAI-compatible tool-call loops	Tested preview + live receipts landed	xaip-openai wrapper
HTTP tools / A2A / proprietary runtimes	Supported receipt flow	xaip-sdk or direct signed receipt emission

The receipt schema is intentionally tool-system-agnostic: agentDid, callerDid, taskHash, resultHash, success, latencyMs, failureType, timestamp. Any agent framework that can hash inputs/outputs and sign with Ed25519 can contribute receipts.

See Emit XAIP receipts from anything for the provider-neutral receipt flow.

Status

v0.4.0 live; v0.5 draft in development (tool class taxonomy + observation/display plumbing).

• v0.5 class metadata plumbing is live for observation/display.

• Class-aware scoring remains a design note and is not used by current trust scores or /v1/select selection behavior.

• Trust Score API (Cloudflare Worker, live)

• Decision Engine (POST /v1/select)

• Aggregator with BFT-capable federation support (public deployment currently single aggregator / quorum:1)

• Ed25519 receipt signing + verification

• Bayesian trust model with caller diversity

• ~3,100 signed receipts across 10 scored MCP servers as of 2026-05-08

• Automated daily data collection (GitHub Actions)

• Published preview receipt producers: xaip-langchain, xaip-openai

• MCP Server: xaip-mcp-trust

• npm: [email protected]

• v0.5 specification draft (tool class taxonomy; class-aware scoring is a design note only)

• Multi-caller diversity mechanism verified (2+ caller identities, metric responds across 8 servers)

• v0.5 class metadata plumbing (observation/display only — does not affect scoring or /v1/select)

• Class-aware scoring (design note only — not live behavior)

• Zero-install caller path: npx xaip-caller (30-second first contribution, demonstrates XAIP beyond MCP)

• External operator callers (mechanism live, external adoption pending — run npx xaip-caller or the full guide)

Writing

• Portable Trust — why trust infrastructure for AI agents must be provider-neutral and behavior-derived (dev.to · Zenn 日本語版)

Related

• xaip-caller — zero-install CLI: npx xaip-caller to contribute to the trust graph
• xaip-mcp-trust — MCP server for AI agents to check trust scores
• xaip-langchain — LangChain.js callback handler that emits XAIP receipts
• xaip-openai — OpenAI tool-calling wrapper with signed receipts
• Veridict — AI agent trust decision layer (runtime monitoring)
• XAIP Specification v0.4 — Current protocol specification
• XAIP Specification v0.5 RC — Release candidate (tool class taxonomy)

License

MIT