1. Conduid
  2. Communication
  3. McpServerManager
MCP server · Communication

McpServerManager

Avalonia desktop app for tracking and inspecting request/session logs with AI assistant

Unclaimed comms
34Low

Scored 6 months ago · breakdown

About McpServerManager

McpServerManager is an MCP server in the Communication category: avalonia desktop app for tracking and inspecting request/session logs with AI assistant.

Install

From source
git clone https://github.com/sharpninja/McpServerManager

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about McpServerManager

Powered by Claude · Grounded in docs

I know everything about McpServerManager. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

McpServerManager

McpServerManager is an Avalonia UI desktop application designed to visualize and analyze session logs and request data from AI coding assistants like GitHub Copilot and Cursor. It provides a unified view of interactions, enabling developers to review prompt history, context usage, and automated actions taken by these tools.

Repository: sharpninja/McpServerManager

Key Features

  • Unified Dashboard: Aggregates logs from different AI providers (Copilot, Cursor) into a single, searchable interface.
  • Tree Navigation: filesystem watcher monitors a target directory for log files, automatically indexing them into a navigation tree.
  • Detailed Request Analysis:
    • Interpretation & Metadata: Displays extracted metadata, user intent interpretation, and key decisions.
    • Context Inspection: View the exact context (code snippets, files) sent to the LLM.
    • Action Tracking: visualizes automated actions (file creation, edits, command execution) in a structured grid.
    • JSON Inspector: Built-in JSON viewer to inspect the raw underlying data for deep debugging.
  • Markdown Rendering: Integrated markdig-based markdown viewer for rendering log content and associated documentation.
  • Responsive UI: Modern, light-theme interface with collapsible sections and persistent window state.

Data Ingestion

The application monitors a specified directory (e.g., docs/requests) for JSON and Markdown log files. It supports custom schema formats used to track AI interactions:

  • Copilot Logs: JSON-based session logs containing request/response cycles, token usage metrics, and workspace context.
  • Cursor Logs: JSON/Markdown hybrid logs capturing "Tab" requests, diffs, and chat history.

Usage

  1. Configure Directory: Point the application to the root folder containing your request logs.
  2. Browse Sessions: Use the tree view on the left to navigate through sessions organized by date or folder structure.
  3. Inspect Requests: Click on a specific request to view its details in the main panel.
    • Expand the Actions grid to see what file changes were performed.
    • Check Interpretation to understand how the AI understood the task.
    • Use the Original JSON expander to see the raw data fields.

Configuration

Edit src/McpServerManager/appsettings.config (copied next to the executable) to set Mcp.BaseUrl, Paths.SessionsRootPath, Paths.HtmlCacheDirectory, and Paths.CssFallbackPath. Session logs are loaded from Mcp.BaseUrl (/mcpserver/sessionlog), while Paths.* is still used for documents/source context and markdown CSS fallback. Paths.HtmlCacheDirectory supports environment variables such as %TEMP%.

Tech Stack

  • Framework: Avalonia UI (Cross-platform .NET XAML framework)
  • Language: C# / .NET 8
  • Parsing: System.Text.Json with custom robust parsing for flexible schemas.
  • Markdown: Markdig.Avalonia for rendering formatted text.

Build and run

# From repo root
cd src/McpServerManager
dotnet run

Deployment automation

Use NUKE from the repo root as the authoritative build/deploy entry point for local automation and CI.

.\build.ps1
.\build.ps1 DeployAll
.\build.ps1 DeployAll --deploy-selection Director,WebUi,DesktopMsix
.\build.ps1 DeployAll --what-if
.\build.ps1 DeployAll --configuration Debug --deploy-selection Director,WebUi
.\build.ps1 BuildAndInstallVsix --what-if
bash ./build.sh VersionInfo --json-output-path artifacts/version.json
bash ./build.sh BuildDesktopDeb --package-version <semver>

Current target names:

  • Director
  • WebUi
  • AndroidPhone
  • AndroidEmulator
  • DesktopMsix
  • DesktopDeb

Behavior notes:

  • build.ps1 and build.sh are the primary entry points; they invoke build\Build.csproj with the repo root wired up for NUKE.
  • When invoked with no arguments, the root wrappers forward --help so you see NUKE help instead of accidentally running a default target.
  • For convenience, the wrappers treat the first bare argument as --target, so commands like .\build.ps1 BuildAndInstallVsix --what-if work without spelling out --target.
  • .github\workflows\build-android.yml is intentionally thin: the workflow restores tools and runner prerequisites, then delegates versioning, packaging, release, F-Droid, and Pages assembly to NUKE targets.
  • The build is best-effort for deploy-all: unavailable targets are skipped and reported in the final summary.
  • --what-if is the standard dry-run mechanism for NUKE-backed targets.
  • DesktopMsix deployment auto-elevates only the certificate trust step through gsudo, then installs the package back in the invoking user context to ensure the app registers for the actual desktop user.
  • DesktopDeb installation on Windows launches an interactive WSL sudo prompt so the user can enter their password when package installation is requested.
  • Legacy files under scripts\ now act as compatibility wrappers so existing commands continue to work while NUKE owns the orchestration logic.
  • For independent target execution, import scripts\DeployAllTargets.psm1 and call the exported compatibility functions directly, for example Invoke-DeployDirectorTool -Configuration Debug -WhatIf.

Validation commands:

  • dotnet build build\Build.csproj -nologo
  • dotnet build src\McpServerManager.Desktop\McpServerManager.Desktop.csproj -nologo
  • Invoke-Pester -Path scripts\DeployAllTargets.Tests.ps1 -Output Detailed -CI

WSL with WSLg

On WSL with WSLg enabled (Windows 11), the app window should appear on the Windows desktop. If it doesn’t:

  1. Check WSLg: Ensure you’re on Windows 11 with WSLg (no separate X server needed).
  2. Run from project: cd src/McpServerManager && dotnet run -c Debug
  3. Or use the script: From repo root, chmod +x run-wslg.sh && ./run-wslg.sh
  4. Taskbar: The window may show in the Windows taskbar; click it to bring it to front.

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About McpServerManager

How do I install McpServerManager?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is McpServerManager safe to use with an AI agent?

Its trust score is 34 out of 100 (low). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is McpServerManager still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.