Project Human

What project-human MCP server does

The project-human MCP server makes nine reusable, rule-based skills available to MCP clients. Its tools cover crisis-signal detection and supportive replies, harmful-content rewriting, WCAG 2.2 accessibility auditing, cognitive accessibility, cultural context, conflict de-escalation, empathetic reframing, neurodiversity-oriented UI review, and age-inclusive design.

The skills are intended for agent workflows that need structured checks or generated guidance around human-centered communication and design. Accessibility audits evaluate pages against the 86 WCAG 2.2 success criteria across A, AA, and AAA levels. Other tools inspect text, user interfaces, or situations against defined patterns and rules.

How it works

The project-human MCP server registers all nine tools with an MCP server factory. Clients discover them through tools/list and invoke them with tools/call. Input validation uses registered schemas, including 18 inline schemas referenced by the project architecture. Handler functions apply local rules and data such as crisis resources, pattern lists, internationalization data, and WCAG criteria.

Processing is rule-based and does not make LLM calls. Skills include explicit safety boundaries, a stated confidence level of low, medium, or high, and automated evaluation gates. The server therefore adds deterministic checks and structured outputs to an agent rather than acting as a separate language model.

Setup and configuration

For a zero-setup connection, configure an MCP client with the hosted endpoint:

{
  "mcpServers": {
    "humanity4ai": {
      "url": "https://humanity4ai.ascent.partners/api/mcp"
    }
  }
}

For local stdio use, run:

npx @humanity4ai/mcp-servers

Then configure the client with command: "npx" and arguments -y and @humanity4ai/mcp-servers. Contributors can clone the repository, install dependencies with pnpm, and start the project. Docker Compose is also supported through docker compose up. Local development requires Node.js 22 or later and pnpm 10 or later. The project lists Windows, macOS, Linux, Android, and iOS among its supported environments.

Tools and capabilities

The nine available capabilities are:

• Detect crisis signals and produce supportive replies with escalation guidance.
• Find harmful, stigmatising, or triggering language and rewrite it.
• Audit web pages against WCAG 2.2 success criteria.
• Review reading level, structure, and cognitive load.
• Check cultural sensitivity for a specified audience and region.
• Create structured, non-coercive de-escalation plans based on conflict intensity.
• Reframe messages with empathy while identifying hollow empathy patterns.
• Review interfaces for ADHD, autism, dyslexia, and sensory-sensitivity considerations.
• Examine user flows for age-related barriers affecting children, adults, and older users.

Limitations and notes

The project-human MCP server describes its skills as non-clinical and rule-based. They do not provide diagnosis, treatment, or professional medical or legal advice. Crisis-related output includes escalation guidance, but the material does not describe the server as a replacement for qualified help.

The repository also provides the same skills through direct prompting and local files, including llms.txt, llms-full.txt, and the skills/ directory. Those options are separate from MCP deployment. The README names VS Code, Cursor, Claude Code, Copilot, Manus AI, OpenCode, OpenClaw, ChatGPT, Claude, and Gemini in its platform and usage guidance; only MCP-compatible clients can use the MCP transport directly.