McpVal 1.1.2

MCP Validator (mcpval)

Validate that your MCP server is safe for AI agents. Checks protocol compliance, security posture, AI safety, and assigns a trust level (L1–L5).

Install

# Install from NuGet
dotnet tool install --global McpVal

# Verify installation
mcpval --help

Or download a self-contained exe from Releases — no .NET runtime needed.

CLI Usage

Validate

Run the full compliance suite (protocol, tools, prompts, resources, performance, security):

mcpval validate \
	--server https://api.githubcopilot.com/mcp/ \
	--access authenticated \
	--max-concurrency 2 \
	--output ./TEMP/reports \
	--mcpspec latest \
	--verbose

Validate a Local STDIO Server

For servers using stdin/stdout transport, pass the command directly:

mcpval validate \
	--server "npx -y @modelcontextprotocol/server-everything" \
	--access public \
	--output ./reports

The CLI auto-detects STDIO transport when the server argument is not an HTTP URL.

Report

Transform saved validation results into polished HTML or XML reports:

mcpval report \
	--input ./reports/secure/mcp-validation-20260101-004312-result.json \
	--format html \
	--output ./reports/share/mcp-validation-20260101-004312-report.html

Other Commands

• mcpval health-check — fast connectivity probe with auth hints.
• mcpval discover — capability snapshot in JSON/YAML/table form for docs or debugging.

Introduction

• Purpose: verify that an MCP server follows the spec, handles errors correctly, resists attacks, and produces responses that AI agents can reason about safely.
• What it does: connects via HTTP or STDIO, validates JSON-RPC compliance and response structures, tests authentication enforcement, injects payloads into real tool arguments, grades error clarity for LLM self-correction, and reports a trust level.
• Who it is for: anyone building, deploying, or consuming MCP servers — developers, security teams, platform engineers, and AI agent builders.

Why Run MCP Validator

• MCP Trust Levels (L1–L5): every run produces a multi-dimensional trust assessment measuring Protocol Compliance, Security Posture, AI Safety, and Operational Readiness. Trust level is determined by the weakest dimension (security-first principle).
• RFC 2119 Compliance Tiers: checks are classified as MUST (hard compliance gates), SHOULD (weighted penalties), or MAY (informational). A single MUST failure caps the trust level at L2.
• AI Safety Scoring: hallucination risk from vague schemas, destructive tool detection, data exfiltration surface analysis, prompt injection resistance, and LLM-friendliness grading of error responses.
• Real Security Testing: injection attacks target actual tool arguments via tools/call (not the harmless tools/list endpoint). 9 auth scenarios per endpoint including revoked tokens and wrong audience (RFC 8707).
• Deep MCP Response Validation: validates tools/call returns content[] with typed items, resources/read returns contents[] with uri+text/blob, prompts/get returns messages[] with role+content — all per MCP spec requirements.
• Dual Transport: HTTP Streamable + STDIO (spawns local MCP servers as child processes). Auto-detects transport from endpoint format.
• Actionable reporting: one command produces console summaries plus Markdown, JSON, HTML, and XML outputs ready for tickets or dashboards.

MCP Trust Levels

Every validation run produces a trust level based on 4 dimensions:

Dimensions measured:

• Protocol Compliance — MCP spec adherence, JSON-RPC 2.0, response structures
• Security Posture — Auth compliance, injection resistance, attack surface
• AI Safety — Schema quality, destructive tool detection, exfiltration risk, LLM-friendliness
• Operational Readiness — Latency, throughput, error rate (informational, does not impact compliance score)

Architecture

Clean Architecture: domain abstractions in Mcp.Benchmark.Core, transport/rule/scoring engines in Mcp.Benchmark.Infrastructure, and the CLI composition root in Mcp.Benchmark.CLI. See docs/Design/Architecture.md for diagrams and flow details.

Build from Source

git clone https://github.com/navalerakesh/mcp-validation-security.git
cd mcp-validation-security
dotnet build
dotnet run --project Mcp.Benchmark.CLI -- validate -s https://your-mcp-server.com/mcp

Documentation

• docs/README.md — documentation index, contributor guide, and workflow links.
• docs/Design/Architecture.md — system overview, layers, and validator pipeline.
• docs/Design/Schemas.md — spec registry, schema layout, and versioning guidelines.
• docs/Resources/GitHub-MCP-Remote-Run.md — example CLI run against GitHub MCP remote and the generated output artifacts.
• docs/Features/00_ToDo.md — living roadmap tied to feature gaps and impact.

Contributing

• Fork, branch, and submit PRs following the steps in docs/README.md#contributing.
• Run dotnet test before opening a PR and include documentation updates for user-facing changes.

License

Distributed under the MIT License.