BinlogInsights

MSBuild binary log (.binlog) analysis tools — a CLI and an MCP server for AI-assisted build investigation.

Why?

MSBuild binary logs capture everything about a build: properties, items, imports, resolved packages, compiler invocations, targets, tasks, and timing. But navigating a binlog is tedious — they can be huge, deeply nested, and full of irrelevant noise.

BinlogInsights exposes structured queries over binlogs through two interfaces:

Interface	Use case
CLI (binlog-insights)	Quick terminal investigation, scripting, CI pipelines
MCP server (binlog-insights-mcp)	AI-assisted investigation in VS Code / Copilot Chat

The MCP server is particularly powerful: point Copilot at a failing binlog and let it drive the investigation — drilling from overview → errors → properties → imports → NuGet diagnostics automatically, following the same workflow a build engineer would.

Features

26 MCP Tools / 13 CLI Commands

Build investigation:

• overview — Build status, duration, project count, top-level results
• errors / warnings — Diagnostics with project, file, line info
• properties — MSBuild property values (with optional filter)
• imports — Full import chain for a project
• items — Item groups (PackageReference, Compile, etc.)
• nuget — NuGet restore diagnostics
• compiler — Full compiler command line (references, defines, flags)
• search — Free-text search across all build messages
• projects — Project list with status and duration
• preprocess — Effective project XML after all imports
• compare — Diff two binlogs: properties, packages, references

Performance analysis (MCP only):

• expensive_projects / project_target_times — Slowest projects and their target breakdown
• expensive_targets / search_targets / project_targets — Target-level timing
• expensive_tasks / search_tasks / tasks_in_target / task_details — Task-level timing
• expensive_analyzers — Roslyn analyzer performance

Evaluation & files (MCP only):

• evaluations / evaluation_properties / evaluation_global_properties — Property/item resolution phase
• list_files / get_file — Access source files embedded in the binlog

Installation

Prerequisites

• .NET 8 SDK or later

Install the MCP server (recommended)

dotnet tool install -g BinlogInsights.Mcp

Install the CLI

dotnet tool install -g BinlogInsights

Build from source

git clone https://github.com/SergeyTeplyakov/BinlogInsights.git
cd BinlogInsights/src
dotnet build
# Install the MCP server as a global tool:
./install.ps1

Setup for VS Code / Copilot

Add the MCP server to your workspace. Create .vscode/mcp.json:

{
  "servers": {
    "binlog-insights": {
      "type": "stdio",
      "command": "binlog-insights-mcp",
      "args": []
    }
  }
}

You can pre-load one or more binlogs at startup so tool calls return instantly:

{
  "servers": {
    "binlog-insights": {
      "type": "stdio",
      "command": "binlog-insights-mcp",
      "args": ["--binlog", "build.binlog", "--binlog", "other.binlog"]
    }
  }
}

Note: Relative paths in --binlog args are resolved against the MCP server's working directory. If files aren't found, use absolute paths or install the Binlog Analyzer extension which sets the working directory automatically.

Once configured, Copilot Chat automatically gains access to all 26 binlog analysis tools.

Setting up your repo for the team

The samples/repo-setup/ directory contains ready-to-copy files. Drop them into your repo root:

your-repo/
├── .vscode/
│   └── mcp.json                          ← MCP server config
├── .github/
│   ├── instructions/
│   │   └── build-investigation.instructions.md  ← auto-activates on build files
│   └── skills/
│       ├── build-tool-setup/
│       │   └── SKILL.md                  ← install/update the MCP tool
│       ├── build-failure-analysis/
│       │   └── SKILL.md                  ← error investigation
│       ├── build-performance-analysis/
│       │   └── SKILL.md                  ← perf bottleneck analysis
│       └── build-comparison/
│           └── SKILL.md                  ← diff two builds

What each file does:

File	Type	Trigger
mcp.json	MCP config	Automatic — VS Code starts the server on workspace open
build-investigation.instructions.md	Instruction	Automatic — activates when you open any .csproj, .sln, .props, .targets, or .binlog file
build-tool-setup/SKILL.md	Skill	On-demand — Copilot loads when the tool isn't installed, install fails, or you need to update
build-failure-analysis/SKILL.md	Skill	On-demand — Copilot loads when you ask about build errors, missing types, NuGet failures
build-performance-analysis/SKILL.md	Skill	On-demand — Copilot loads when you ask about slow builds, build duration, expensive targets
build-comparison/SKILL.md	Skill	On-demand — Copilot loads when you compare builds, diff environments, or migrate build systems

A SKILL-TEMPLATE.md is also included as a starting point for creating your own skills.

Team members just need to:

1. Install the tool: dotnet tool install -g BinlogInsights.Mcp
2. Open the repo in VS Code — everything else is automatic

Usage

Generate a binlog

Build your project with binary logging enabled:

dotnet build /bl:build.binlog

Or for MSBuild:

msbuild /bl:build.binlog

CLI examples

# Build overview
binlog-insights overview build.binlog

# Show errors
binlog-insights errors build.binlog

# Check properties for a specific project
binlog-insights properties build.binlog --project MyApp --filter TargetFramework

# Search build messages
binlog-insights search build.binlog --query "PackageReference"

# Compare two builds
binlog-insights compare before.binlog after.binlog

# NuGet restore diagnostics
binlog-insights nuget build.binlog

# Compiler command line
binlog-insights compiler build.binlog --project MyApp

AI-assisted investigation

With the MCP server configured, ask Copilot Chat:

"My build failed. Investigate build.binlog and tell me what went wrong."

Copilot will automatically:

1. Call overview to check build status
2. Call errors to see what failed
3. Drill deeper based on error type:
   • Missing types → items to check PackageReference
   • Property issues → properties to inspect values
   • Import problems → imports to trace the chain
   • NuGet failures → nuget for restore diagnostics
4. Use search for free-text investigation
5. Use compare to diff two builds

"Why is my build slow? Analyze build.binlog performance."

Copilot will:

1. Call expensive_projects to find the slowest projects
2. Drill into targets with project_target_times
3. Check expensive_analyzers for slow Roslyn analyzers
4. Identify bottleneck tasks with expensive_tasks

"I have two binlogs from different build configurations. Compare them and tell me what's different."

Copilot will:

1. Call compare to diff properties, per-project packages, and solution-wide packages
2. Report property differences (e.g., NuGetPackageRoot, SDK paths)
3. Show per-project package version mismatches
4. List solution-wide package diffs — packages resolved to different versions across all projects
5. Drill into specifics with properties, items, or compiler to explain why versions differ

This is especially useful for:

• Migrating build systems (e.g., CoreXT → stock MSBuild) — verify that the new build resolves the same packages
• Debugging CI vs local — find property or package differences between environments
• Before/after changes — confirm a .props or .targets change had the intended effect

Architecture

BinlogInsights.Core        — Query engine (MSBuild.StructuredLogger)
├── BinlogInsights          — CLI tool (System.CommandLine)
└── BinlogInsights.Mcp      — MCP server (ModelContextProtocol SDK)

• Core parses binlogs via MSBuild.StructuredLogger and exposes typed query classes
• CLI wraps queries with System.CommandLine for terminal use
• MCP wraps queries as Model Context Protocol tools for AI agents

License

MIT