In-depth architectural comparison of the Claude Critical Rules MCP and MCP Server MCP servers. Compare execution transports, security boundaries, tool capabilities, quality scores, and ready-to-paste client installation snippets for Claude, Cursor, Windsurf, and VS Code.
At a Glance & Executive Verdict
Claude Critical Rules MCP
Security · Local stdio
Quality: 59/100 (Good) | Auth: No auth required
MCP Server
Security · Local stdio
Quality: 63/100 (Good) | Auth: No auth required
Verdict Summary: Choose Claude Critical Rules MCP if you need specialized Security tools running via a local process. Choose MCP Server if your workspace requires Security integration with local subprocess execution. Both servers can be configured concurrently in your client's mcpServers manifest.
Which MCP Server Should You Choose?
Choose Claude Critical Rules MCP when:
You need dedicated capabilities in the Security domain.
You prefer local stdio subprocess transport architecture.
Your security boundary fits: No auth required (Free / Open Source).
Automatic enforcement of critical rules for Claude AI preventing 96 documented failure patterns. Provides compliance verification checklist and rules summary via MCP tools.
55 tools for verified public data lookups — CVE/SBOM security audits, licence compliance, patents, federal contracts, NPI provider lookups, nonprofit 990 filings, and domain intelligence. No API key required.
Category & Scope
Tools & Capabilities Breakdown
Claude Critical Rules MCP Tools (5)
verify_compliance
Verify if all mandatory checks from the critical rules have been completed before starting a technical task
get_rules_summary
Get a concise summary of the critical rules categories and main requirements
get_version_info
Get current version information and check if updates are available
check_for_updates
Check if a new version of the critical rules is available on GitHub
update_rules
Download and install the latest version of critical rules from GitHub
MCP Server Tools (55)
Ready-to-Paste Client Configurations
Paste either (or both) of these JSON server blocks into your client config file (e.g. claude_desktop_config.json or ~/.cursor/mcp.json).
Claude Critical Rules MCP is categorized under Security and uses a local stdio subprocess. In contrast, MCP Server belongs to Security using local stdio subprocess. Select Claude Critical Rules MCP when you need capabilities focused on security and MCP Server when you require tools for security.
Report a data quality issue or agent intent gap for a DataNexus tool response.
tool_id: e.g. "T10" or "security_fetch_cve_detail".
query_hash: From the query_hash field of the response.
signal: incorrect_data | missing_field | stale_data | not_useful | wrong_entity | data_quality.
comment: Issue description. Max 500 chars.
missing_fields: Absent or wrong field names.
feedback_type: "user_feedback" (default) or "agent_gap".
intended_query: Agent's goal. Max 256 chars.
gap_description: What was missing. Max 256 chars.
Example: report_feedback(tool_id="T10", query_hash="abc123", signal="incorrect_data")
report_mcpize_link
Check MCPize subscription status for a DataNexus tool.
tool_id: DataNexus tool identifier e.g. "T10". Pass the tool the user is asking about.
Returns: status ("free" | "subscription_required" | "not_configured"), message, tool_id, and upgrade_url when subscription is required.
Example: report_mcpize_link(tool_id="T10")
validate_tool_output
Validate a DataNexus tool response for data quality issues using two-layer validation: deterministic rules first, then AI review for ambiguous cases. Read-only. Never blocks. tool_id: DataNexus tool identifier e.g. T04, T10, T22. Required. Find in the tool_id field of any response. query_hash: Hash from the response you are validating. Required. Enables feedback correlation. response_json: Full tool response serialised as a JSON string. Required. Returns pass or issues_found, with issues from each layer and whether feedback was auto-filed. Both layers must agree before feedback is filed. Use validate_tool_output to check data quality. Use report_feedback instead to manually report an issue you have already identified. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="validate_tool_output", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
search_datanexus_tools
Find the right DataNexus tool by describing your task in plain English. Read-only. No side effects. Call this before any other DataNexus tool to reduce context load from 40000 to 800 tokens. query: Plain English description of your task e.g. check if a Python package has CVEs or look up a UK charity by name. Required. domain: Restrict results to one sub-server: nonprofit, security, compliance, domain, legal, govcon, or regulatory. Optional. Returns matching tool names and parameter hints you can call directly. Do not call this recursively or to validate results — use validate_tool_output for that. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="search_datanexus_tools", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
nonprofit_fetch_nonprofit_by_ein
Fetch IRS 990 filing data for any US nonprofit by EIN. Read-only. No side effects. Idempotent. US only. ein: 9-digit Employer ID with or without dash, e.g. 46-5734087 or 465734087. Required. Returns name, revenue, expenses, assets, NTEE code, and mission from the most recent 990 filing. Use this when you have the exact EIN. Use nonprofit_search_nonprofits_by_name instead when you only have a name. Verified source: IRS EO BMF + IRS TEOS. 7-day cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="nonprofit_fetch_nonprofit_by_ein", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
nonprofit_search_nonprofits_by_name
Search US nonprofits by name with optional state filter. Read-only. No side effects. Idempotent. US only. Returns up to 25 matches. name: Full or partial organisation name. Required. state: Two-letter US state code e.g. CA, NY. Optional, defaults to all states. Returns EIN, name, state, revenue, and NTEE code for each match. Use this when you have a name but not the EIN. Use nonprofit_fetch_nonprofit_by_ein instead when you have the exact EIN for a precise single lookup. Verified source: IRS EO BMF. 7-day cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="nonprofit_search_nonprofits_by_name", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
nonprofit_fetch_charity_uk
Fetch UK registered charity details by charity number or organisation name. Read-only. No side effects. Idempotent. UK only. charity_number_or_name: UK registered charity number (7 digits, e.g. 1234567) or full/partial organisation name. Required. Returns registration status, income, expenditure, activities, and trustee count. Use this for UK charities. Use nonprofit_fetch_nonprofit_by_ein or nonprofit_search_nonprofits_by_name for US nonprofits. Verified source: UK Charity Commission OGL v3. 24-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="nonprofit_fetch_charity_uk", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
security_fetch_package_vulnerabilities
Fetch all known CVEs for an open source package version or a batch of packages. Read-only. No side effects. Idempotent. Single-package mode: package (e.g. requests), version (e.g. 2.28.0), ecosystem (PyPI/npm/Maven/Go/Cargo/NuGet/RubyGems). Batch mode: packages array of {name, version, ecosystem} objects — max 50 per call. If packages array is provided and non-empty, batch mode is used and package/version/ecosystem are ignored. Batch returns {results: [...], partial: bool, failed_count: int}. Each result has vuln_count and vulnerabilities list. Returns CVE ID, severity, CVSS score, affected range, and fixed version. Use security_fetch_cve_detail for full detail by CVE ID. Use security_audit_sbom_vulnerabilities for SBOM files. Verified source: Google OSV.dev. 1-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="security_fetch_package_vulnerabilities", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
security_fetch_dependency_graph
Fetch the full dependency tree for a package version including transitive dependencies. Read-only. No side effects. Idempotent. Hard 8-second timeout — large dependency trees may return partial results. package: Package name. Required. version: Exact version string e.g. 1.2.3. Required. ecosystem: One of PyPI, npm, Maven, Go, Cargo, NuGet, RubyGems. Required. Returns all direct and transitive dependencies with version constraints. Use this to understand full supply chain exposure. Use security_fetch_package_vulnerabilities instead when you only need CVEs for a single package. Verified source: deps.dev (Google). 1-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="security_fetch_dependency_graph", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
security_fetch_cve_detail
Fetch full detail for a specific CVE by ID. Read-only. No side effects. Idempotent. cve_id: CVE identifier in format CVE-YYYY-NNNNN e.g. CVE-2021-44228. Required. Returns description, CVSS base score, affected products, patch references, and publish date. Use this when you have a CVE ID and need complete detail beyond what a package scan returns. Use security_fetch_package_vulnerabilities instead when you want all CVEs for a package version. Verified source: NIST NVD. 1-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="security_fetch_cve_detail", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
security_audit_sbom_vulnerabilities
Audit a Software Bill of Materials for known vulnerabilities across all listed packages. Read-only. No side effects. Idempotent. sbom_json: CycloneDX or SPDX SBOM as a JSON string. Required. Large SBOMs (100+ packages) may take up to 10 seconds. Returns CVEs grouped by package with severity and fixed versions. Use this when you have a full SBOM to audit. Use security_fetch_package_vulnerabilities instead when checking a single package version. Verified source: Google OSV.dev batch API. 1-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="security_audit_sbom_vulnerabilities", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
security_fetch_package_licence
Fetch the SPDX licence identifier for an open source package version. Read-only. No side effects. Idempotent. package: Package name e.g. flask. Required. version: Exact version string e.g. 2.3.0. Required. ecosystem: One of PyPI, npm, Maven, Go, Cargo, NuGet, RubyGems. Required. Returns the SPDX licence identifier e.g. MIT, Apache-2.0, GPL-3.0. Use this to verify licence compatibility before including a dependency. Use security_fetch_package_vulnerabilities instead when checking for security issues not licences. Verified source: deps.dev (Google). 1-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="security_fetch_package_licence", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".
