APCore: Build once, invoke by Code or AI
A schema-enforced module standard for the AI-Perceivable era.
The Problem
Today's software is a "black box" to AI. Developers spend countless hours writing custom JSON schemas, manual mapping code, and bespoke MCP servers just to let Claude or an OpenAI agent call a simple function. Then they repeat the process for every new protocol. It's a fragmented, unmanaged maintenance nightmare.
The Solution: apcore
apcore is an AI-Perceivable module standard that makes your code naturally perceivable and understandable by AI through enforced Schema definitions and behavioral annotations. Build your module once — it works as a regular function call, an MCP tool, an A2A agent skill, an OpenAI tool, or a CLI command. Zero code changes between them.
Core Architecture
Schema Enforcement: input_schema, output_schema, and description are mandatory. No schema = no registration. This is what makes modules AI-perceivable by default.
4 Integration Paths: Start from scratch with Class-based modules, use a 1-line @module decorator, or use apcore-toolkit for zero-code migration of legacy systems.
Progressive Disclosure: Three-layer metadata reduces token bloat by up to 90% in large toolsets.
Self-Healing Agents: ai_guidance in errors provides Agents with recovery strategies to fix inputs autonomously.
Enterprise Pipeline: 11-step execution engine with pattern-based ACL, Middleware, and W3C tracing.
Build Once — The Adapter Ecosystem
Write your module once. Zero-config adapters handle the rest:
apcore-mcp: MCP Server + OpenAI Tools provider with browser-based Tool Explorer UI
apcore-a2a: A2A agent with auto-generated Agent Cards and streaming
apcore-cli: Schema-driven CLI with STDIN piping and approval gates
apcore-toolkit: Scans Django/Flask/Express routes → generates apcore bindings
Governance Ready (NIST 2026 RFI)
apcore is one of the first module standards built to address the 2026 U.S. Federal Government RFI on AI Agent Security:
Blast Radius Control: Pattern-based, default-deny ACL enforces Least Privilege.
Safety Gates: Mandatory human-in-the-loop for destructive or requires_approval operations.
Full Traceability: W3C-compliant tracing for every agent action across your stack.
Multi-Language SDKs
Available in Python (pip install apcore), TypeScript (npm install apcore-js), and Rust (cargo add apcore) — with identical semantics.
Stop writing glue code. Build once, invoke by Code or AI.
GitHub: https://github.com/aiperceivable
Docs: https://apcore.aiperceivable.com/
Replies