Skip to content
MCP server

Saferagenticai MCP

Read-only access to 239 agentic AI safety patterns and 15 operational heuristics

Install
Pricing
Open source
Platforms
CLI, self-hosted, API
Runs
Locally or Remotely
Version
0.4.2
Listing updated
today

Similar MCP servers

About Saferagenticai MCP

Saferagenticai MCP is an open-source MCP server for security. It can run locally on your machine or as a hosted remote server.

This server exposes the Safer Agentic AI framework, a catalogue of safety criteria grouped into 16 suites, plus an implementation patterns layer, so a coding assistant can look up guidance while it designs or reviews an agent. The registry describes 239 patterns and 15 heuristics, while the README text gives 238 pattern files and 14 heuristics for the packaged release, and notes that early wheels bundled fewer.

Twelve read-only tools cover the corpus. You can list suites and requirements with filters, fetch one requirement with its pattern, search patterns with field-weighted ranking, resolve a partial ID, and follow citations in either direction. A task-based tool takes a natural-language description and returns the top patterns grouped by suite. Others report review coverage and list operational heuristics drawn from production agentic AI deployments.

It runs over stdio from PyPI or uvx, or as a hosted streamable HTTP endpoint. The code is MIT licensed and the framework content is CC-BY-4.0, credited to Nell Watson and the Agentic AI Safety Community of Practice. The README deliberately keeps the server read-only, and pinning a version is advised for audit reproducibility.

Install

Run in a terminal

claude mcp add --transport http saferagenticai-mcp https://mcp.saferagenticai.org/mcp

Add to Settings → Connectors → Add custom connector

https://mcp.saferagenticai.org/mcp
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "saferagenticai-mcp": {
      "url": "https://mcp.saferagenticai.org/mcp"
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "saferagenticai-mcp": {
      "type": "http",
      "url": "https://mcp.saferagenticai.org/mcp"
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "saferagenticai-mcp": {
      "serverUrl": "https://mcp.saferagenticai.org/mcp"
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "saferagenticai-mcp": {
      "type": "http",
      "url": "https://mcp.saferagenticai.org/mcp"
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Tools this server provides (12)

Taken from the project documentation; the running server may differ.

list_suites
Lists the 16 suites with titles and subgoal counts.
get_requirement
Returns one subgoal with its pattern layer.
list_requirements
Lists subgoals filtered by suite, type, content type or confidence.
search_patterns
Ranks pattern matches for a query.
get_cross_references
Lists outgoing adjacencies for an ID.
get_reverse_references
Lists which patterns cite a given pattern.
resolve_id
Canonicalises a partial ID, slug fragment or display ID.
find_patterns_for_task
Returns top patterns grouped by suite for a task description.
list_unreviewed
Lists patterns without a reviewer, lowest confidence first.
review_stats
Reports review coverage and validation issue counts.
list_operational_heuristics
Lists operational heuristics, optionally filtered by suite or keyword.
get_operational_heuristic
Returns one operational heuristic by ID.

What it does

Pattern search
Ranks patterns by weighted matches on title, summary and body.
Task-based lookup
Returns top patterns grouped by suite for a described task.
Cross-references
Follows citations outgoing and incoming between patterns.
Requirement browsing
Lists suites and filters requirements by suite, type or confidence.
Operational heuristics
Provides heuristics distilled from production agentic AI deployments.
Review statistics
Reports review coverage per suite and confidence level.

Good for

  • Find safety patterns relevant to an agent you are designing
  • Look up the full text of a framework requirement by ID
  • Trace which patterns cite a given pattern
  • Check review coverage of the framework
  • Pull operational heuristics into a design review

Pricing

Open source

Code MIT; framework content CC-BY-4.0

Saferagenticai MCP FAQ

Can it modify the framework?

No. It is read-only and deliberately has no write tool.

How many tools does it have?

Twelve, including search_patterns, find_patterns_for_task and get_requirement.

What licences apply?

The server code is MIT, and the framework content is CC-BY-4.0.

How do I install it?

Use uvx, pipx or pip with the saferagenticai-mcp package, or connect to the hosted endpoint.

Is Saferagenticai MCP free?

Yes. Saferagenticai MCP is open source, so you can use and self-host it for free.

How do I add Saferagenticai MCP to Claude Code?

Run this command in your terminal: claude mcp add --transport http saferagenticai-mcp https://mcp.saferagenticai.org/mcp

What are the best alternatives to Saferagenticai MCP?

The closest alternatives to Saferagenticai MCP are REA, headroom, omniget, next-ai-draw-io and mcp-for-blender.

Where this comes from

  • Description: written from the project's README.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers