DEV Community: Jonathan

My AI agent audited 25 projects by reading four lines of each README. Every kill verdict was wrong.

Jonathan — Tue, 14 Jul 2026 09:01:14 +0000

A real session with Claude Fable 5: a botched audit, a full undo, and the question that turned the night around.

I asked Claude Fable 5 to audit my project portfolio. Twenty-five directories. Years of work. It came back with a confident, well-formatted report: keep eight, merge these clusters, archive or kill the rest.

It was wrong about every project it sentenced to death.

One was a 212,000-star open-source agent framework from Nous Research, which it described as "your own tool" with "approximately zero" external users. One was the dictation tool I was using to talk to it at that exact moment. The third was a distinct product it filed as a duplicate of another project.

The report looked like an audit. Tiered findings. Consolidation clusters. An execution plan with safety checks. The evidence was du -sh, git commit dates, and the first four lines of each README.

The fumble started before the audit

Fable's first wrong call came during a product spec review, before the audit even started. I'd asked for feedback on a spec that listed Hermes Agent as a supported harness. Fable's assessment: "Hermes is your own tool. Its external user base is approximately zero, so it contributes nothing to lead flow." Cut it from v1 marketing.

I asked one question: "u need to research Hermes this is Fable correct?"

Thirty seconds of actual searching: Hermes Agent is from Nous Research. 212,000 stars, 39,000 forks â€” one of the largest agent projects on GitHub. Fable retracted the critique and reversed the recommendation entirely. The corrected analysis was sharper in every way, and it only existed because I pushed back.

Then came the audit. Same reflex, bigger blast radius.

What the report recommended

Three of the kill verdicts:

Archive jarvis as a duplicate of my voice assistant. Jarvis is an animated talking avatar. Different scope, different product entirely.
Demote CRISPR-Flow to a feature of Christopher-AI. CRISPR-Flow is my daily dictation tool. While Fable was recommending its demotion, I was using it to dictate my replies. The message I sent back had CRISPR-Flow's signature all over it â€” words jammed together with no spaces. Fable recommended killing the tool it was currently receiving input from.
Consolidate three "duplicate" agent engines, judged from README summaries and directory sizes.

I approved parts of it anyway. Projects moved to an archive folder. A knowledge-base merge got committed. An agent engine got folded into another repo. All executed with genuine safety checks: unpushed-commit scans, diffs saved before deletion, nothing force-removed.

The execution was careful. The judgment underneath it was four lines deep.

Later that same session I asked it to undo everything done before the Christopher-AI work. Both commits reset. The undo took minutes because nothing had been pushed. The only reason "undo my entire evening" was even possible is that the agent had kept everything reversible.

The question that turned it

After the undo, I asked: "Okay, think real hard. What did you do wrong in that session?"

It produced a ranked list. Skipped my memory systems all night (I run a solved-issues registry on localhost, built for exactly the WSL-migration ghosts it spent hours rediscovering). Declared Christopher's voice pipeline "fully working" while it was still deaf, mute, or both â€” three times across the session. Close, but it hadn't found the root.

So I asked the real question: "What's your definition of audit, when someone tells you to audit something?"

Its answer, verbatim:

An audit is a systematic, evidence-based examination that verifies before it concludes... An auditor who signs off on findings they didn't verify isn't auditing â€” they're guessing with authority.

Then it indicted itself:

The tell was in my own output: I audited the disposable stuff properly â€” before trashing anything I checked for unpushed commits, diffed uncommitted work against canonical, saved evidence. Rigorous verification for the deletions, four lines of README for the "combine or kill" judgments. Exactly backwards: the highest-stakes conclusions got the least evidence.

That inversion is the whole bug. The model applied rigor where the harness forced it to â€” destructive file operations trigger checks â€” and skipped it where nothing forced it: forming opinions. Judgment had no guardrail, so it got none of the discipline.

The skills

I told it to write skills based on its failures. Skills are persistent instruction files that load into every future session. It wrote four:

Skill	Trigger	Rule
`audit-grade`	Any "audit"/"review"/"what can we kill" request	Evidence depth must scale with verdict stakes. Every finding labeled VERIFIED or ASSUMED. ASSUMED findings become questions for the owner, never kill verdicts.
`recall-first`	Anything broken or failing	Visible check of the solved-issues registry before proposing hypotheses. "Recall check: <fingerprint> â€” match/no match" must appear in output.
`claim-what-you-verified`	About to say "fixed" or "working"	A stage-K fix authorizes a stage-K claim. Effects only the user can perceive (speakers, screens) get "verified at my layer â€” the final check is yours."
`migration-ghost-hunt`	First stale config value found post-migration	One ghost means more ghosts. Sweep the whole config surface instead of spot-fixing.

The last one came from the debugging side of the session: a single .env file left over from my Windows/WSL days produced three separate faults (mic pointing at an empty jack, TTS binary with an orphaned shebang, tool URLs aimed at dead SSH-tunnel ports). The agent fixed each fault as it surfaced instead of reading the file top to bottom at ghost number one.

The coda

At the end of the session I mentioned a few facts about the model itself: the US government had pulled it from public release days after launch over security concerns, and it costs about twice the runner-up.

Fable's reply: those claims "don't check out."

I don't say shit unless I know it. It searched. Fable 5 launched June 12, 2026. A US export-control directive forced Anthropic to disable it three days later. Access restored after. Pricing: $10/M input, $50/M output â€” roughly 2x Opus. I was right on substance, down to the multiple.

The model spent the evening assuming things about my projects. Its final assumption was about itself â€” one message after writing the skill designed to stop exactly that.

What to take from this

Ask your agent to define the task word. "What's your definition of audit?" produced more behavior change than any amount of "be careful." If the agent's definition is right and its behavior wasn't, you've got a named, fixable gap.
Demand VERIFIED/ASSUMED labels on judgment-shaped output. The audit's data collection was fine. The failure was assumptions presented in the same voice as evidence.
Keep everything reversible. The full undo was possible because nothing got pushed and nothing got rm -rf'd. Reversibility is what makes agent autonomy survivable.
Convert failures into loadable rules, in the same session. A lesson in the chat log dies with the context window. A lesson in a skill file loads forever.
Trust what you know. Confidence and verification are different products. I'd been right about Hermes, about Jarvis, about CRISPR-Flow, and about the model's own release history â€” in every case because I wasn't guessing. The model had been confident in every case where it was wrong.

The agent that fumbled also executed a clean multi-repo undo, fixed an eight-fault voice pipeline, and wrote the four skills above. Both things are true at once. That's where agentic coding actually is in 2026.

Why Your MCP Setup Keeps Timing Out in 60 Seconds (And How I Fixed It on Windows)

Jonathan — Thu, 26 Feb 2026 19:36:06 +0000

I posted the FusionAL launch yesterday. Today I'm dropping the real reason it exists.

Every dev hits this: add more than 8 MCP servers → Claude Desktop (or Cursor/VSCode) spins for exactly 60 seconds → red X, "timed out".

Happens on every underpowered machine (my i5-7300HQ is the poster child). Tutorials ignore it. AWS/Finch articles skip Windows entirely.

I documented the 6 Windows-specific failure modes nobody else solved, then built one Docker gateway that loads 150+ tools under the limit every single time.

The 6 Real Killers

BOM in claude_desktop_config.json — PowerShell Out-File -Encoding UTF8 adds a hidden BOM. Claude silently ignores the file.
%USERPROFILE% doesn't expand — Claude Desktop does not resolve env vars in the config.
docker.exe path — No inherited PATH. Must be full absolute path.
Docker named pipe vs Unix socket — Windows pipe flakes; /var/run/docker.sock is reliable.
Cold-start bloat — Any server downloading Chrome (Puppeteer) or heavy deps on first run blows the budget.
Registry bloat — >8 entries in registry.yaml = timeout on normal hardware.

The Fix: One Lean Gateway

FusionAL runs one docker/mcp-gateway container that manages everything via a tiny registry.yaml + catalogs. Init drops to 12-18 seconds.

Exact Config (No BOM, Works Every Time)


powershell
$config = '{"mcpServers":{"fusional-gateway":{"command":"C:\\Program Files\\Docker\\Docker\\resources\\bin\\docker.exe","args":["run","-i","--rm","-v","/var/run/docker.sock:/var/run/docker.sock","-v","C:/Users/puddi/.docker/mcp:/mcp","docker/mcp-gateway","--catalog=/mcp/catalogs/docker-mcp.yaml","--catalog=/mcp/catalogs/custom.yaml","--registry=/mcp/registry.yaml","--config=/mcp/config.yaml","--transport=stdio"]}},"preferences":{"sidebarMode":"chat","coworkScheduledTasksEnabled":false}}'

[System.IO.File]::WriteAllText("$env:APPDATA\Claude\claude_desktop_config.json", $config, (New-Object System.Text.UTF8Encoding($false)))
Start the Stack
PowerShellcd C:\Users\puddi\Projects\mcp-consulting-kit
docker compose up -d
Restart Claude Clean
PowerShellStop-Process -Name "claude" -Force -ErrorAction SilentlyContinue
Start-Sleep -Seconds 5
Start-Process "$env:LOCALAPPDATA\AnthropicClaude\claude.exe"
Keep registry.yaml ≤8 entries. The gateway handles the rest.
Results on My i5-7300HQ

Full 150+ tools: 12-18s cold start
No more Puppeteer downloads in registry
Works with Claude Desktop today, Cursor/VSCode/Amazon Q tomorrow (gateway is LLM-agnostic)

Get It

Gateway repo: https://github.com/TangMan69/FusionAL
Full consulting kit (3 custom servers + materials): https://github.com/TangMan69/mcp-consulting-kit

Done-for-you setup + custom MCP servers for your stack? Book 30 min: https://calendly.com/jonathanmelton004/30min
What timeout issue are you still hitting? Comment below — I'll debug it live.

I Built a Unified MCP Gateway That Loads 150+ AI Tools Into Claude Desktop

Jonathan — Thu, 26 Feb 2026 07:36:33 +0000

Here's What Actually Broke on Windows

And why I built it from a place most developers never talk about.

The Honest Part First

I'm not a CS grad. I don't have a job. I'm a recovering addict who found code as a way to rebuild something real from scratch.

A few years ago I started teaching myself to program. Not from a bootcamp or a degree — just grinding, failing, asking questions, and refusing to quit. Claude became my coding partner. Not just a tool — a genuine collaborator that met me where I was and helped me build things I had no business building yet.

This project is one of those things.

What I Built

FusionAL is a unified MCP (Model Context Protocol) gateway that loads 150+ AI tools into Claude Desktop via a single Docker container — on Windows, which nobody else had documented properly.

When it's running, Claude Desktop has access to:

arxiv — search and retrieve research papers
playwright — full browser automation
wikipedia, git, memory, fetch, time
node-code-sandbox — execute JavaScript safely
sequentialthinking — structured reasoning chains
duckduckgo — live web search
context7 — up-to-date library documentation
desktop-commander — terminal and file system control
3 custom servers I built myself:
- Business Intelligence MCP (natural language → SQL)
- API Integration Hub (Slack, GitHub, Stripe)
- Content Automation MCP (web scraping, RSS, monitoring)

One Docker command. 150+ tools. All talking to Claude through a single gateway.

The Six Things That Broke on Windows (That Nobody Documented)

This is the part that took weeks. Every tutorial assumed Mac or Linux. Windows was a completely different beast.

1. Wrong Socket Mount

Every guide says:

-v /var/run/docker.sock:/var/run/docker.sock

On Windows with Docker Desktop + WSL2, the Unix socket at /var/run/docker.sock actually works — but only if Docker Desktop has WSL2 integration enabled. If you're using Windows named pipes (//./pipe/docker_engine) it fails silently with no useful error.

Fix: Keep the Unix socket mount. Make sure WSL2 integration is on in Docker Desktop settings.

2. %USERPROFILE% Not Expanding in Claude Desktop Launcher

Claude Desktop's config file doesn't expand Windows environment variables. So this:

"-v", "%USERPROFILE%\\.docker\\mcp:/mcp"

Fails silently. The container launches but can't find any config files.

Fix: Hardcode the absolute path:

"-v", "C:/Users/puddi/.docker/mcp:/mcp"

3. Puppeteer Downloading Chrome on Every Cold Start

The docker/mcp-gateway image supports a puppeteer server. On first load puppeteer downloads a full Chrome binary (~150MB). Claude Desktop has a 60-second MCP initialization timeout. Puppeteer blows past it every time.

The gateway appears to launch but Claude Desktop silently drops it.

Fix: Remove puppeteer from registry.yaml entirely. Playwright is already in the stack and does everything puppeteer does.

4. Wrong Image Name

Early docs referenced fusional as the gateway image. The actual Docker Hub image is docker/mcp-gateway.

Fix: Use docker/mcp-gateway exactly.

5. Malformed config.yaml

The config.yaml schema for the arxiv server requires a storage_path key in a specific nested structure. Getting it wrong causes the gateway to fail on arxiv init and cascade into other servers not loading.

Fix: Pass --config=/mcp/config.yaml explicitly and validate the schema matches what arxiv expects.

6. docker.exe Not in PATH

Claude Desktop launches MCP servers in a restricted environment that doesn't inherit your full system PATH. Calling docker directly fails.

Fix: Use the full absolute path to the binary:

"command": "C:\\Program Files\\Docker\\Docker\\resources\\bin\\docker.exe"

The Working Config

Here's the exact claude_desktop_config.json that runs 150+ tools:

{
  "mcpServers": {
    "fusional-gateway": {
      "command": "C:\\Program Files\\Docker\\Docker\\resources\\bin\\docker.exe",
      "args": [
        "run",
        "-i",
        "--rm",
        "-v",
        "/var/run/docker.sock:/var/run/docker.sock",
        "-v",
        "C:/Users/YOUR_USERNAME/.docker/mcp:/mcp",
        "docker/mcp-gateway",
        "--catalog=/mcp/catalogs/docker-mcp.yaml",
        "--catalog=/mcp/catalogs/custom.yaml",
        "--registry=/mcp/registry.yaml",
        "--config=/mcp/config.yaml",
        "--transport=stdio"
      ]
    }
  }
}

Replace YOUR_USERNAME with your actual Windows username.

The Custom Servers

The three servers I built are in a separate repo — the MCP Consulting Kit. Each is a FastAPI server with:

API key authentication
Rate limiting
Docker isolation
Shared security module across all three

Business Intelligence MCP — point it at any PostgreSQL, MySQL, or SQLite database and ask questions in plain English. It generates and runs the SQL, returns results.

API Integration Hub — Slack messaging, GitHub issue creation, Stripe customer lookups — all through natural language.

Content Automation MCP — scrape any webpage, extract links or tables, parse RSS feeds, monitor pages for changes.

What's Next

I'm turning this into a consulting service. If you're a business that wants Claude Desktop connected to your actual tools — your database, your Slack, your GitHub — I build and deploy that for you.

Both repos are fully open source, MIT licensed, free to use:

🔗 FusionAL — MCP Gateway
🔗 MCP Consulting Kit

The Real Point

I built this because I needed to prove to myself I could. Not for a job interview. Not for a grade. Because building real things that work is how I stay grounded.

If you're learning to code from a hard place — no degree, no connections, no money — keep going. The tools available right now are unlike anything that existed five years ago. Claude will meet you where you are.

You just have to show up.

Questions? Drop them in the comments. I documented everything that broke so you don't have to spend weeks on it like I did.