Labsco
yfedoseev logo

Fossil MCP

★ 62

from yfedoseev

The code quality toolkit for the vibe coding era.

🔥🔥🔥🔥✓ VerifiedFreeAdvanced setup
#automation#devops#git#security

Fossil MCP

The code quality toolkit for the vibe coding era.

Static analysis that finds the mess vibe coding leaves behind — dead code, duplicated logic, scaffolding artifacts, and disconnected functions — across 16 languages.

fossil-mcp.com

The Problem

AI-assisted coding is fine — you review the code, you understand the architecture, you stay in control. Vibe coding is different. You describe what you want, the AI writes it, and you ship it without reading every line. Tools like Claude Code, Cursor, GitHub Copilot, and Windsurf make this workflow fast and productive. But over days and weeks, vibe-coded projects accumulate a specific class of problems that traditional linters don't catch:

Dead code piles up fast. When the AI refactors a function, it writes the new version but often forgets to remove the old one. You don't notice because you didn't read the diff line by line. Over multiple sessions, unused functions, unreachable branches, and orphaned utilities pile up — the codebase grows but nothing gets pruned. A METR study found developers spend significant time checking and debugging AI output. Dead code makes this exponentially harder.

Duplication spreads silently. Each AI session has limited context window. It generates a utility function that already exists elsewhere, or solves the same problem with a slightly different implementation three files over. You asked for a feature, it works, so you move on. Traditional duplicate detection focuses on copy-paste — vibe coding duplication is structural: similar logic, different names, scattered across modules.

// Phase 1, // TODO, // Step 2 — everywhere. AI agents work in phases. They leave behind scaffolding markers that were meant to be temporary: // Phase 1: Setup, // TODO: implement error handling, placeholder function bodies with pass or todo!(), and phased naming like process_data_v2. In vibe coding, nobody goes back to clean these up. They become permanent fixtures.

Functions exist that nothing calls. This is the vibe coding signature. The AI writes a helper function, uses it, then in a later session rewrites the caller to use a different approach — but the helper stays. Without a call graph, neither you nor the AI can tell which functions are actually connected to the rest of the codebase. Current AI coding tools navigate code by text search, not by understanding how functions call each other.

Temp files accumulate in the repo. AI sessions create temp_, backup_, old_, phase_1_ files and directories. In vibe coding, you don't audit your file tree after each session. These artifacts persist across commits.

The Solution

Fossil MCP is a static analysis toolkit purpose-built for vibe-coded projects. It detects the artifacts that accumulate when AI writes most of the code — and it works both as a CLI tool for developers and as an MCP server that gives AI agents a code graph instead of just text search.

  ███████╗ ██████╗ ███████╗███████╗██╗██╗           ()    ()
  ██╔════╝██╔═══██╗██╔════╝██╔════╝██║██║            \    /
  █████╗  ██║   ██║███████╗███████╗██║██║             |  |
  ██╔══╝  ██║   ██║╚════██║╚════██║██║██║             |  |
  ██║     ╚██████╔╝███████║███████║██║███████╗       /    \
  ╚═╝      ╚═════╝ ╚══════╝╚══════╝╚═╝╚══════╝      ()    ()
  Dig up dead code. Unearth clones. Expose scaffolding.

What Fossil Detects

AnalysisWhat it findsThe vibe coding problem
Dead CodeUnreachable functions, unused exports, orphaned methodsAI rewrites a caller but forgets to delete the old helper — nobody notices
Code ClonesType 1 (exact), Type 2 (renamed), Type 3 (structural) duplicatesEach AI session reinvents utilities that already exist elsewhere in the codebase
ScaffoldingPhase N / Step N comments, TODO/FIXME markers, placeholder bodiesAI works in phases and leaves temporary markers that never get cleaned up
Temp Filestemp_*, backup_*, old_*, phase_* files and directoriesSession artifacts that persist because nobody audits the file tree
Code GraphTrace paths between any two functions, blast radius analysis, call graph traversalAI tools navigate by text search — Fossil gives them a graph to trace how functions connect and what breaks if you change one

What Makes Fossil Different

  • Purpose-built for vibe coding. Not a general linter — specifically targets the mess that accumulates when AI writes most of the code and humans review less of it.
  • Graph, not grep. AI coding tools navigate code by searching for text. Fossil builds a call graph and lets agents trace how functions connect, find blast radius before refactoring, and discover dead ends — without reading every file.
  • MCP-native. Runs as an MCP server so AI agents can self-check their output during development.
  • Saves tokens, saves money. Instead of an agent scanning files over and over to find issues, Fossil identifies dead code, clones, and scaffolding in one pass — fewer rounds of LLM inference, lower cost.
  • Built in Rust. Single binary, no runtime dependencies. Scans thousands of files in seconds. Memory-safe by design.
  • Cross-file analysis. Resolves imports, barrel re-exports, and class hierarchies to find dead code across module boundaries.
  • Framework-aware. Auto-detects React, Next.js, Django, Spring, Axum, and more — won't flag lifecycle methods as dead code.
  • Zero configuration. Works out of the box. Config file is optional.
  • 16 languages. One tool for polyglot codebases.

Supported Languages

LanguageExtensions
Python.py
JavaScript.js, .jsx, .mjs
TypeScript.ts, .tsx
Rust.rs
Go.go
Java.java
C#.cs
C/C++.c, .h, .cpp, .cc, .cxx, .hpp
Ruby.rb
PHP.php
Swift.swift
Kotlin.kt
Scala.scala
Bash.sh, .bash
R.r, .R

CI/CD Integration

Fossil includes a check command for CI/CD pipelines. It fails builds when code quality thresholds are exceeded, helping teams enforce code standards and prevent technical debt from accumulating.

Basic Usage

# Check against configured thresholds
fossil-mcp check

# Override thresholds via CLI
fossil-mcp check --max-dead-code 10 --max-clones 5

# Diff-aware mode (only analyze changed files in PR)
fossil-mcp check --diff origin/main

# Generate SARIF for GitHub code scanning
fossil-mcp check --diff origin/main --format sarif

# Quiet mode (no diagnostic output)
fossil-mcp check --quiet

Configuration

Add a [ci] section to fossil.toml:

[ci]
max_dead_code = 10           # Maximum dead code findings (0 = fail on any)
max_clones = 5               # Maximum clone findings
max_scaffolding = 3          # Maximum scaffolding findings
min_confidence = "medium"    # Minimum confidence (low|medium|high|certain)
fail_on_scaffolding = false  # Fail if any scaffolding found

GitHub Actions Integration

Create .github/workflows/fossil-check.yml:

name: Fossil CI Check

on:
  pull_request:
  push:
    branches: [main]

jobs:
  fossil:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Install Fossil
        run: curl -fsSL fossil-mcp.com/install.sh | sh

      - name: Run Fossil check
        run: |
          fossil-mcp check \
            --diff origin/${{ github.base_ref || 'main' }} \
            --format sarif \
            > fossil-results.sarif

      - name: Upload to GitHub Security
        if: always()
        uses: github/codeql-action/upload-sarif@v3
        with:
          sarif_file: fossil-results.sarif

How It Works

  1. Scans the project using the same analysis engine as scan
  2. Optionally filters to only changed files (via --diff branch)
  3. Evaluates against configured thresholds
  4. Reports findings as text, JSON, or SARIF
  5. Exits with code 1 if thresholds exceeded (fails CI build)

Exit Codes

CodeMeaning
0All thresholds passed ✓
1Threshold exceeded (build fails)
2Error (missing git, invalid config, etc.)

For complete examples, see examples/fossil.toml and examples/fossil-check.yml.

How It Works

  1. Scan — walks project files, respects .gitignore, skips vendored/generated code
  2. Parse — builds tree-sitter ASTs for each source file (16 languages)
  3. Extract — pulls functions, calls, imports, attributes, and class hierarchy from ASTs
  4. Graph — builds a cross-file CodeGraph with import resolution and barrel re-export support
  5. Analyze — detects entry points (via heuristics + framework presets), runs reachability analysis, identifies dead code and clones
  6. Report — outputs findings as text dashboard, JSON, or SARIF