Apex Protocol – An open MCP-based standard for AI agent trading
Article URL: https://apexstandard.org/ Comments URL: https://news.ycombinator.com/item?id=47655618 Points: 4 # Comments: 1
v0.1.0-alpha
APEX defines how AI agents communicate with brokers, exchanges, dealers, and other execution venues. One protocol. Realtime state. Autonomous safety. Multi-asset by design.
FIX Protocol connected electronic trading. APEX connects agentic trading.
Protocol Specification
19 mandatory tools across five domains — session, account, orders, market data, and risk — plus a realtime resource model with live state, notifications, and production capability profiles.
Read the spec
Instrument Taxonomy
A universal instrument identifier system. APEX:FX:EURUSD means the same thing at every broker. No more per-broker symbol mapping. One ID scheme for every tradeable instrument.
See the ID format
Conformance Tests
167 executable conformance checks across 4 reference implementations validating protocol conformance — from basic tool smoke tests to transport resilience, schema validation, and production readiness.
View the test suite
AI Agent
APEX Protocol (MCP over HTTP/SSE)
Broker MCP Server
Tools for actions. Resources for live state. Notifications for change. The broker implements it. The agent speaks it. No hub. No routing. No middleman.
For Brokers
Implement once. Every agent connects.
Build an APEX alpha-compatible MCP server and become accessible to agent integrations through a shared protocol surface — without bespoke integrations for each one.
- Reference implementations in TypeScript, Rust, Go, and Java
- HTTP/SSE transport with session replay and notification delivery
- 167-check conformance harness across stdio and transport
- Canonical instrument IDs across all asset classes
- Production profiles: Realtime and Autonomous capability claims
Broker integration guide
For Agent Developers
Write once. Trade on any broker.
Use canonical instrument IDs and a unified tool vocabulary. Your agent code targets a shared protocol surface rather than per-provider APIs. Switch providers with far less custom integration work.
- Canonical instrument IDs: APEX:FX:EURUSD
- Realtime state: subscribe to quotes, candles, positions, risk
- Event-driven notifications: fills, rejections, candle closes, kill switch
- Deterministic safety: halt on stale data, broken sequences, risk limits
- Open spec, free to implement: CC-BY 4.0
Agent quickstart
Core
Layer 1 — Mandatory Baseline
Session, account, orders, market data, risk
v0.1-alpha
Production
Realtime & Autonomous Profiles
Transport, resources, notifications, safety controls
v0.1-alpha
Profile
FX
Spot FX, CFD FX, rollovers, currency exposure
v0.1-alpha
Profile
CFD
Equities, indices, commodities, corporate actions
v0.1-alpha
Profile
Crypto
Spot, perpetuals, funding rates, margin modes
v0.1-alpha
Profile
Derivatives
Listed options, futures, greeks
planned
Profile
Fixed Income
Bonds, yield, duration
planned
apex.session.* Authentication, capabilities, heartbeat*
apex.account.* Balances, positions, orders, history*
apex.order.* Place, modify, cancel, status*
apex.market.* Quotes, snapshots, search, instrument details*
apex.risk.* Pre-trade checks, account limits*
Live State Resources
Quotes, candles, positions, orders, risk, and derived features exposed as MCP resources with freshness metadata and monotonic sequencing. Subscribe once, stay current.
Structured Notifications
Seven mandatory notification types — order fills, partial fills, rejections, candle closes, kill switch, gap fill, and replay failure — each with a typed envelope carrying sequence, timestamp, and payload.
HTTP/SSE Transport
Streamable HTTP on a single /mcp endpoint. Session management via Mcp-Session-Id. SSE delivery with monotonic event IDs and Last-Event-ID reconnect.
Session Replay
Acknowledgment-driven event log replays execution-critical events on reconnect. Ephemeral market data collapsed into gap fill markers. Agent controls retention via apex.session.acknowledge. Storage is implementation choice — memory, file, or durable queue.
Autonomous Safety
Kill switch, position limits, daily loss caps, stale-data rejection, and sequence-gap detection. The runtime halts autonomy before the model is asked to decide — not after.
Production Profiles
Two capability claims beyond the core baseline: Production Realtime for live trading and Production Autonomous for agent-driven execution with full safety controls.
Open standard. Vendor-neutral. Community-governed.
The APEX Protocol specification is licensed CC-BY 4.0. Reference implementations are Apache 2.0. Governance is through the Technical Advisory Committee with an open RFC process. Agent developers never pay to use the protocol.
Sign in to highlight and annotate this article

Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
agent
Architecture and Orchestration of Memory Systems in AI Agents
The evolution of artificial intelligence from stateless models to autonomous, goal-driven agents depends heavily on advanced memory architectures. While Large Language Models (LLMs) possess strong reasoning abilities and vast embedded knowledge, they lack persistent memory, making them unable to retain past interactions or adapt over time. This limitation leads to repeated context injection, increasing token [ ] The post Architecture and Orchestration of Memory Systems in AI Agents appeared first on Analytics Vidhya .
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Self-Evolving AI

Show HN: hot or not for .ai websites
I originally took all domains crawled by CommonCrawl and made my own private tool for exploring the public web. I've used it to get some sense/intuition for what's being made by others. It's helped me avoid proposing the 100th agentic marketing SaaS, when brainstorming with friends :) So thought others might see the fun in exploring websites this way. I've tried my best to circumvent the Cookie popups for the pictures, and have removed most parked domains, access denied, bad gateway etc. I've also filtered out subdomains, as there's too much noise. Lmk if there's anything else about it you'd like to know. Only the frontend and two basic pg tables have been made specifically for this project. The rest have been part of other hobby projects, e.g. parsing CC indexes (and WET data), taking scr





Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!