MCP agent-readable instructions

MCP helps agents discover BRIK64 docs, command contracts, PCD rules, and evidence workflows through a structured surface.

  • CLI command contract: Current command name, availability, install path, and allowed command families.
  • PCD authoring rules: Syntax, inputs, outputs, monomer selection, and fail-closed behavior.
  • Evidence vocabulary: How to report PCD hash, command output, checked scope, missing evidence, assumptions, and handoff status.
  • Managed-section rules: Agent instruction writes must be explicit, reviewable, and user-approved.
  • Boundary: MCP provides structured context. It does not approve code, skip CLI checks, or create evidence by itself.

MCP

Agent-readable context for bounded logic work.

MCP helps agents discover BRIK64 docs, command contracts, PCD rules, and evidence workflows through a structured surface.

MCP docsAgent skills

Agent Integration

MCP should make the current BRIK64 rulebook easy for agents to read.

The goal is not autonomous authority. The goal is accurate context, bounded actions, and clear handoff.

Agent Resources

The MCP surface should expose the same BRIK64 objects reviewers use.

Agents should not infer commands, syntax, or public claims from stale memory.

agent context

CLI command contract

Current command name, availability, install path, and allowed command families.

PCD authoring rules

Syntax, inputs, outputs, monomer selection, and fail-closed behavior.

Evidence vocabulary

How to report PCD hash, command output, checked scope, missing evidence, assumptions, and handoff status.

Managed-section rules

Agent instruction writes must be explicit, reviewable, and user-approved.

Boundary: MCP provides structured context. It does not approve code, skip CLI checks, or create evidence by itself.

Agent Handoff

Every agent report should name the BRIK64 object it touched.

The handoff format keeps agent work auditable without depending on visual page layout.

structured report

Resource used

Which docs, command contract, or skill source informed the work?

Files and PCDs changed

Which PCD files, .brik workspace metadata, registry references, or agent sections changed?

Evidence status

What was verified, what failed, and what remains assumed?

Hosted MCP behavior depends on the workspace where it is enabled.

Give agents current context before they shape logic.

Use MCP docs with the public agent skill so generated work starts from bounded BRIK64 rules.

Read MCP docsAgent docs