Assay.Mcp 0.2.1

Assay

Code analysis for AI agents. A .NET CLI that builds a queryable graph from a C# solution — types, members, packages, dependencies, health signals — and emits structured markdown or JSON sized for LLM context windows.

Built so an AI coding agent can investigate a large C# codebase the way a human would: orient first, then drill down. Equally useful at the terminal for humans triaging an unfamiliar repo.

Install

dotnet tool install -g Assay

Requires the .NET 10 SDK.

Quick start

Run from any directory containing a .sln or .slnx:

assay overview            # Solution structure, project table, health summary
assay health              # Errors, failing tests, security findings
assay vulns               # NuGet vulnerability scan
assay search OrderService # Fuzzy search across the codebase

Auto-detects a single solution file in the current directory. Pass a path explicitly for any other case (assay overview path/to/Solution.slnx).

What it does

Parses a C# solution into an immutable directed graph (CodeGraph) where every code element is a vertex and every relationship is an edge:

.sln/.slnx → structural parse → Roslyn enrichment → CodeGraph → queries → markdown/JSON

Graph contents: solutions, projects, namespaces, types, members, files, assemblies, packages, vulnerabilities, traits — connected by 10 relation types (References, Inherits, Implements, Contains, DependsOn, Tests, UsesPackage, HasVulnerability, Calls, Overrides). Calls/Overrides give a method-level call graph.

Caching: Results are cached by source hash in a per-user cache directory (ASSAY_CACHE_DIR, else $XDG_CACHE_HOME/assay or ~/.cache/assay on Unix/macOS, %LocalAppData%/assay on Windows). Subsequent runs skip Roslyn analysis entirely. --no-cache forces a fresh build.

How it differs from existing tools

• dotnet list package / dotnet sln list — flat lists. Assay returns a queryable graph with semantic relations and health signals fused in.
• Roslyn analyzers — produce diagnostics at compile time. Assay produces answers on demand: "what depends on this type", "which methods are highest CRAP risk", "what's the dependency path between A and B".
• IDE call hierarchies — interactive only. Assay is scriptable and emits structured output an agent or pipeline can consume.

Commands at a glance

Add --json for machine-readable output. Add --max-tokens N to size output to an LLM token budget. Names resolve by short/partial/full qualification; exit codes 0/2/64/66/70.

Sample output

$ assay vulns

# Vulnerabilities: SampleShop

**Packages scanned:** 5 | **Vulnerable:** 1

## High (1)
| Package | Version | Advisory | Fix Version | Projects |
|---------|---------|----------|-------------|----------|
| Newtonsoft.Json | 11.0.2 | https://github.com/advisories/GHSA-5crp-9r3c-p9vr | 13.0.1 | SampleShop.Infrastructure |

For AI coding agents

Assay is designed as a tool for AI coding agents. It ships three Claude Code Skills — assay-code-analysis, assay-test-coverage, and assay-dependencies — so the agent automatically reaches for assay when navigating a C# codebase:

assay skills install          # Install to ~/.claude/skills/
assay skills install --local  # Install to ./.claude/skills/

Or wire it in as an MCP server (Claude Code and other MCP clients):

{ "mcpServers": { "assay": { "command": "assay", "args": ["mcp"] } } }