Codeward MCP Server

An MCP server that gives AI coding agents first-class knowledge of Codeward. Instead of guessing at policy syntax or a CVE's severity, the agent calls a tool and gets an authoritative answer.

It runs in two modes:

• Offline (default) — policy authoring, validation, explanation and registry tools, served from a knowledge bundle embedded in the binary. No network, no credentials.
• Intel-connected — set CODEWARD_INTEL_API to additionally expose live vulnerability, package health, licence compatibility and supply-chain data.

No API key required

The MCP server only reads Intel's anonymous endpoints — packages, vulnerabilities, licences, CWEs, ecosystem signals, EOL, stats. It does not use the key-gated bulk scanner query endpoint, so Intel-connected mode works without a token.

Install

Not publicly distributed yet

There is no public codeward-io/mcp repository or release, so the server cannot currently be downloaded. Email [email protected] if you want access while this is being sorted out.

# Download for your platform, then:
chmod +x codeward-mcp

Connect a client

The server speaks MCP over stdio (default) or HTTP. Most desktop assistants launch it over stdio.

Claude Desktop / Claude Code

{
  "mcpServers": {
    "codeward": {
      "command": "/absolute/path/to/codeward-mcp",
      "env": {
        "CODEWARD_INTEL_API": "https://intel.codeward.io"
      }
    }
  }
}

Omit CODEWARD_INTEL_API to run fully offline.

What the agent can do

Policy authoring — the offline half. get_policy_type, get_rule_type, get_operator, get_action, get_output_config, get_example, find_rule_type, find_field, get_docs_reference, get_scanner_info. These let an agent write a valid .codeward.yaml without inventing syntax.

Policy review — analyze_policy_coverage finds gaps in a config; diff_policies explains what changed between two versions; suggested_policies proposes a starting set.

Vulnerability intelligence — lookup_vulnerability, get_vulnerability_by_package, search_vulnerabilities, get_exploit_status, explain_cwe, explain_owasp_category, get_reachability_hints.

Package and supply chain — lookup_package, search_packages, get_version_info, get_supply_chain_trust, check_typosquatting, suggest_alternative, get_upgrade_impact, get_ecosystem_signals, check_runtime_eol.

Licences — check_license_compatibility, compare_licenses, list_licenses.

Working with scan results — explain_finding, explain_finding_batch, explain_scan_output, recommend_remediation, assess_dependency, audit_dependencies.

Validation and profiles — validate_policy, explain_policy, policy_drift_check, get_profile, suggest_profile (the registry profiles).

Browsing Intel — search_intel, get_intel_stats, get_recent_activity, get_package_detail.

Keeping it in step with the scanner

The embedded knowledge bundle describes the scanner's config surface. When the scanner's configuration changes, the bundle is regenerated so the agent does not emit stale policy syntax — if you build the server from source after a scanner release, run make update-registry first.

Privacy

In offline mode the server makes no network calls at all. In Intel-connected mode it reads public Intel endpoints. It never sends your source code anywhere.