March 2026Product

Axiom Refract Ships Native MCP Server for AI Agent Integration

16 tools via Streamable HTTP — Claude Code, Cursor, Windsurf, and any MCP-compatible agent can query architectural records directly

FOR IMMEDIATE RELEASE

Cerebrum Cadre today announced that Axiom Refract ships with a native MCP (Model Context Protocol) server, making it one of the first architecture governance platforms to provide direct AI agent connectivity to a codebase's structural record. The MCP server exposes 16 tools over Streamable HTTP transport, enabling any MCP-compatible client to query dependency graphs, blast radius calculations, SPOF analysis, compliance mappings, and remediation guidance programmatically.

What MCP Is and Why It Matters

Model Context Protocol is an open standard for connecting AI agents to external data sources and tools. Rather than feeding an AI model raw files and asking it to infer structure, MCP allows the model to query a verified data source and receive structured answers. The distinction matters: inference is probabilistic; a governed architectural record is deterministic.

Axiom Refract's MCP server gives AI coding tools — including Anthropic's Claude Code, Cursor, Windsurf, and any client implementing the MCP specification — the ability to ask questions about a codebase's architecture and receive answers derived from the platform's AST-parsed, graph-analyzed structural data.

The 16 Tools

The MCP server exposes 16 purpose-built tools organized around the architectural governance workflow:

  • System overview and danger zone identification — compressed structural maps, SPOF flags, and risk summaries
  • File-level analysis — centrality metrics, blast radius, complexity scores, and dependency chains for any file
  • Graph operations — node queries with sorting and filtering, edge traversal, and cluster identification
  • Zone-based navigation — architectural zone lookup, file grouping, entry point identification, and per-zone risk assessment
  • Migration and maintenance — prioritized refactoring plans, safe-to-remove dead code lists, and ghost method detection
  • Evidence and compliance — reasoning chains behind analysis findings, database topology, and supply chain audit data

Transport: Streamable HTTP

Axiom Refract's MCP server uses Streamable HTTP as its transport layer — the current recommended transport in the MCP specification. Clients connect via standard HTTPS POST requests to a single endpoint. Session management is handled via the mcp-session-id header. CORS is fully enabled for browser-based clients.

What AI Agents Never See

Axiom Refract's MCP server exposes structural metadata — not source code. AI agents querying the platform receive dependency graphs, centrality scores, risk assessments, and compliance mappings. They do not receive function bodies, raw source files, or proprietary business logic. All data served through MCP uses PII-redacted structural representations.

No customer code processed through Axiom Refract is used to train AI models. This commitment is documented in the platform's Terms of Service, Privacy Policy, and Data Processing Agreement.

Availability

The MCP server is available now to all Axiom Refract subscribers. Configuration requires a single endpoint URL. Documentation and client configuration examples for Claude Code, Cursor, and Windsurf are available at axiomrefract.com.

About Cerebrum Cadre

Cerebrum Cadre, a division of CSI, Inc., builds tools for teams managing architectural complexity at scale. Axiom Refract is the first commercial product — SKU 1 — an architecture governance platform that produces a universal, machine-readable record of any codebase's structure, risk, and compliance posture. Cerebrum Cadre is currently in formation as an independent corporation, targeting mid-2026.

© 2026 Cerebrum Cadre, a division of CSI, Inc.