Capabilities

Explore the capabilities shared by Mecatl's clients and servers. These guides explain how each capability works, how to configure it, and which deployments support it.

Get oriented

• Capability and deployment matrix explains the operational differences among server and terminal deployments.
• What is a cloud-native harness? explains how Mecatl separates the agent process from durable state and execution.
• Ask Mecatl about itself covers the account of Mecatl that every agent carries and how it answers questions about the product.

Run and maintain sessions

• The agent loop explains turns, events, tool dispatch, permission pauses, and terminal states.
• Start and resume sessions covers creating, selecting, and continuing sessions.
• Choose models and providers covers provider, model, and reasoning-effort selection.
• Model routing assigns models to internal calls and delegated tasks, and explains routing decisions.
• Multimodal input covers image and other content blocks supported by the selected provider.
• Context windows covers model context-window resolution and overrides.
• Session continuity covers persistence, recovery, retention, and maintenance.
• Scheduled tasks covers recurring and one-shot runs.
• Core tools describes the built-in catalog and execution model.

Configure agent behavior

• Subagents, teams, and parallel work compares the delegation tools and their workspace behavior.
• Define named agents covers agent definitions and specialist delegation.
• Skills, commands, and soul covers reusable instructions and persona.
• Project instructions and rules covers trusted project guidance and .claude/rules.
• Memory and user model explains project memory, the user model, and their lifecycle tools.
• Learning covers evidence-backed learning, reflection, and learned-skill admission.
• Dreaming and memory consolidation covers periodic consolidation and manual review.

Secure connections and execution

• Permissions and posture covers permission modes, trust, guardrails, and autonomous operation.
• Caller identity and OIDC covers authenticated caller identity and ownership separation.
• MCP OAuth and credentials covers OAuth profiles and credential storage.
• MCP client covers transports, tool names, resources, prompts, and reconnect behavior.
• Execution environments covers workspaces, shells, forks, and environment persistence.
• Hook system covers lifecycle hooks that observe, block, or transform agent activity.

Understand runtime behavior

• Observability and resilience explains runtime signals, model-call recovery, and product metrics.
• Collect metrics, traces, and diagnostics covers operator collection and endpoint configuration.