As AI-driven agents and automated workflows become increasingly widespread, structured, predictable interactions between intelligent agents and external systems are critical. Currently, integrations among agents, code, and external services typically require custom coding, proprietary formats, or hardcoded schemas, resulting in duplicated effort, brittle interactions, and poor interoperability.
Specifically, these challenges include:
-
Agent ⇄ Agent Communication:
No widely adopted standard exists for structured interactions between autonomous AI agents, forcing ad-hoc, incompatible integration patterns. -
Agent ⇄ Code Communication:
While semi-standards like OpenAI’s Function Calling have emerged to facilitate structured agent-to-code communication, adoption remains partial and tied closely to specific vendor implementations. A more universally compatible, open approach would improve robustness and interoperability across diverse agents and tooling ecosystems. -
Code ⇄ Agent Communication:
External code calling into intelligent agents typically involves custom, poorly structured APIs, causing unclear integration points, inconsistent results, and challenging error handling.
Existing API standards (such as OpenAPI) address structured data schemas but aren’t specifically optimized for these dynamic, schema-driven agent interactions. A more streamlined, agent-centric approach is needed.
The Structured Completion Protocol (SCP) provides a lightweight, schema-driven standard designed explicitly for clear, predictable interactions among agents, code, and external services. It defines a simple HTTP-based pattern:
- Discovery (GET): Agents retrieve a JSON schema defining exactly what structured inputs are required.
- Submission (POST): Agents submit structured data adhering strictly to the provided schema.
- Structured Response: Responses have clear, explicit schemas, simplifying integration and reducing ambiguity.
Additionally, SCP recommends an optional standardized asynchronous callback pattern for handling long-running tasks, ensuring consistent, predictable asynchronous interactions.
SCP deliberately aligns with and complements existing standards to ensure broad compatibility:
- JSON Schema & OpenAPI: SCP leverages JSON Schema definitions directly, making it immediately familiar to API developers accustomed to OpenAPI.
- AI Tooling Standards: SCP aligns naturally with emerging AI frameworks (e.g., OpenAI Function Calling, LangChain tools), building upon these existing semi-standards to provide a broader, vendor-neutral structured interaction model.
- Web & Hypermedia APIs: SCP complements existing hypermedia-driven standards (HAL, JSON Hyper-Schema), ensuring seamless integration with modern web APIs.
SCP also complements the Model Context Protocol (MCP)—an emerging standard specifically designed to standardize secure, structured data retrieval for AI agents. While MCP focuses primarily on secure context access and data sources, SCP complements MCP by defining structured interactions and requests between agents, tools, and services.
In other words:
- MCP standardizes secure data access and retrieval for AI agents.
- SCP standardizes structured interactions, requests, and completions.
Used together, SCP and MCP provide a complete, robust, and standardized interaction and data-access ecosystem for AI-driven agents.
The intention behind SCP is to establish a simple yet robust open standard specifically optimized for structured agent interactions. SCP aims to reduce integration complexity, improve interoperability, and encourage ecosystem-wide adoption by providing clearly structured, schema-driven communication patterns.
SCP’s minimalistic and flexible design ensures it can adapt to evolving requirements. The ultimate vision is a cohesive ecosystem where intelligent agents, code, and services can seamlessly discover, invoke, and interact with each other, enabling more efficient, reliable, and intelligent automation across diverse domains.