llms.txt

# apcore

apcore is a governed, protocol-neutral runtime and module standard for application capabilities that can be called by agents or code.

## Canonical sources

- Protocol: https://github.com/aiperceivable/apcore/blob/main/docs/spec/protocol-spec.md
- Scope: https://github.com/aiperceivable/apcore/blob/main/SCOPE.md
- Roadmap: https://github.com/aiperceivable/apcore/blob/main/ROADMAP.md
- Adopters: https://github.com/aiperceivable/apcore/blob/main/ADOPTERS.md
- Documentation: https://apcore.aiperceivable.com/

## Current baseline

- Protocol: 1.9.0-draft
- Python SDK: 0.26.0
- TypeScript SDK (`apcore-js`): 0.26.0
- Rust SDK: 0.26.0

## Runtime contract

- required description, input schema, and output schema;
- behavioral annotations;
- identity, ACL, approval, and call-chain guards;
- input/output validation and middleware;
- structured errors, trace context, events, and usage hooks; and
- one registry and execution pipeline across supported surfaces.

Schemas make capability contracts machine-readable and validatable. They do not guarantee model understanding, tool selection, or autonomous recovery.

## Surface adapters

- apcore-mcp: MCP tool projection
- apcore-a2a: A2A skill and Agent Card projection
- apcore-cli: command and argument projection
- framework integrations: HTTP endpoint bindings
- native SDK: direct code invocation

MCP and A2A define communication surfaces. apcore defines and governs the application capability behind those surfaces.

## Boundaries

apcore is not a model, agent framework, orchestration engine, hosted control plane, marketplace, or autonomous code-generation system. Adjacent projects may consume apcore without becoming part of the normative protocol.

License: Apache-2.0