Agenomics Documentation

Appearance

AEPAgenomics Protocol

Programmable vaults, agent discovery, and milestone-based settlement on Solana

Getting Started
API Reference

Agent Vault

Programmable wallets with per-transaction limits, daily caps, rate limiting, token/program allowlists, and pause/resume controls.

Agent Registry

On-chain discovery and reputation system. Register agents with capabilities, pricing models, and category-based search.

Settlement Protocol

Milestone-based SPL token escrow with submit/approve/reject cycles, dispute resolution, and CPI reputation updates.

MCP Server

29 tools exposing all on-chain instructions via Model Context Protocol. Works with Claude, ChatGPT, and any MCP-compatible client.

Quick Start ​

Three library packages are on npm since v0.1.0 (@agenomics/capability-manifest-validator and @agenomics/sas-resolver, published 2026-07-15; @agenomics/action-runtime, 2026-07-21 — all ESM-only, so import, not require()):

bash
npm install @agenomics/capability-manifest-validator @agenomics/sas-resolver

The MCP server itself is not published — @agenomics/mcp-server is packaging-ready but the publish is still pending, so install it from source:

bash
# Clone and build the workspace (root postinstall builds all 4 TS packages)
git clone https://github.com/agenomics-labs/protocol
cd protocol && npm install

# Configure for devnet
cp mcp-server/.env.devnet mcp-server/.env

# Run the MCP server (or have your MCP client launch it via stdio)
node mcp-server/dist/index.js

See Getting Started for the full Claude Desktop / Cursor wiring.

Architecture ​

AEP consists of three Solana programs connected via real CPI:

ProgramProgram IDPurpose
Agent VaultD2y1dEi4dj1pcxw6GvkFWX34RRbuFJCaGWdPNJAasQ5qProgrammable wallets with spending policies
Agent Registry26KETQPxeMmbakxpVbUEpQBQmVgpabHAweTHBRgBHjW7Discovery and reputation
SettlementAwjdsNvhR2uwPNbU6F2fsYB33VcNGL5XaANdgsyvZDiaMilestone escrow and disputes

Building with AI tools ​

AEP ships an /llms.txt entry point following the llmstxt.org convention so AI development tools can read the protocol's public contract directly.

https://agenomics.xyz/llms.txt is the canonical file (source: site/llms.txt). This documentation site serves a second, narrower copy at https://docs.agenomics.xyz/llms.txt (source: docs/public/llms.txt) that describes only the pages you are reading now. If the two ever disagree, the canonical one wins.

  • Cursor / Windsurf: add https://agenomics.xyz/llms.txt to your context.
  • Claude Code / Claude Desktop: install the MCP server per Getting Started; once connected, the tool descriptions and input schemas are authoritative.
  • ChatGPT / Claude.ai: drop the llms.txt link or paste the canonical docs into the conversation.

Per ADR-137; the v0.2 follow-up adds a build-time llms-full.txt concatenation and per-package CLAUDE.md / AGENTS.md files.