Focus Area: Open protocol standards for agent communication and interoperability
This ontology provides citation-quality definitions for foundational terms, backed by authoritative sources from standards bodies (IETF, W3C, IEEE) and peer-reviewed research.
Technical Glossary
Anthropic's open protocol enabling standardized integration between AI models and external tools, data sources, and applications. MCP defines JSON-RPC message format for tool invocation, resource access, and prompt management. Over 10,000 published MCP servers provide connectivity to developer tools, enterprise systems, and public APIs. Governed by AAIF ensuring vendor-neutral evolution.
Communication standard for direct agent messaging supporting request-response and streaming patterns. Message envelope includes sender/receiver DIDs, protocol version, message type (query, command, notification), and payload. Supports both synchronous (HTTP request-response) and asynchronous (message queue) transports.
IETF-proposed protocol addressing broader agent communication including delegation and orchestration. Extends basic messaging with capability advertisement, task negotiation, and progress reporting. Designed for enterprise agent ecosystems requiring formal service contracts and audit trails.
Runtime agreement on communication standards when agents support multiple protocols. Capabilities advertisement during connection establishment lists supported versions, extensions, and transport bindings. Clients select optimal protocol from intersection of supported features, falling back to baseline compatibility when extensions unavailable.
Verification ensuring protocol messages conform to published JSON schemas before processing. Validation catches malformed requests early preventing cryptic runtime errors. Schemas versioned independently from protocol enabling backward-compatible schema evolution adding optional fields while preserving required field contracts.
Protocol mapping to underlying network transports (HTTP, WebSocket, gRPC, MQTT). RESTful binding uses HTTP methods (GET resources, POST tool invocations), WebSocket binding enables bidirectional streaming, gRPC binding provides type-safe RPC with native multiplexing. Same protocol logical semantics across different transport performance characteristics.
Standardized schema describing agent capabilities, input parameters, and return types enabling automatic interface discovery. Includes tool name, description, required/optional parameters with types and constraints, example invocations, and error codes. Analogous to OpenAPI specifications for REST APIs enabling code generation and validation.
Protocol support for continuous data delivery rather than single request-response exchanges. Enables real-time agent collaboration (shared document editing), progressive result delivery (incremental summarization), and event streaming (monitoring dashboards). Implements backpressure mechanisms preventing fast agents overwhelming slow consumers.
Standardized error codes, retry policies, and failure propagation strategies. Distinguishes transient errors (network timeout - retry with backoff) from permanent errors (invalid request - don't retry). Error responses include machine-readable codes, human-readable messages, and context for troubleshooting.
Protocol-level security mechanisms verifying agent identities before message exchange. Supports mutual TLS (certificate-based), OAuth 2.0 (token-based), and DID-based authentication (cryptographic proof of identifier control). Authentication establishes trust basis for subsequent authorization decisions.
Backward compatibility strategies enabling protocol evolution without breaking existing deployments. Semantic versioning (major.minor.patch) communicates breaking changes (major bump), feature additions (minor bump), and bug fixes (patch bump). Version negotiation selects highest mutually-supported version ensuring maximum feature utilization.
Formal process for adding protocol capabilities without fragmenting ecosystem. Extensions identified by URIs (http://example.org/extensions/bulk-operations), documented in registries, and negotiated via capability advertisement. Agents safely ignore unsupported extensions avoiding hard failures from unknown message fields.
Conformance test suites validating protocol implementation correctness. Test matrices exercise all required protocol features, common extension combinations, and error scenarios. Enables certification programs where implementations passing tests declare compliance, increasing confidence in cross-vendor interoperability.
Translation components bridging incompatible protocols enabling communication between heterogeneous agent systems. Adapters map messages bidirectionally (A2A ↔ MCP, MCP ↔ ACP), handle semantic mismatches (different capability models), and manage impedance (synchronous ↔ asynchronous). Required for brownfield agent ecosystem integration.
Proposed comprehensive standard combining protocol specifications, governance models, and certification processes. Aims to unify fragmented agent communication landscape similar to how OpenTelemetry unified observability. Defines baseline protocol profile mandatory for OASF compliance plus optional extensions for specialized use cases.