APOLOGY-AS-A-SERVICE MCP Server

Apology-as-a-Service (AaaS)

"Because sometimes, 'my bad' just isn't enough."

🟢 Live endpoints:

• Streamable HTTP (current MCP): https://apology-as-a-service-production.up.railway.app/mcp
• SSE (legacy, kept for older clients): https://apology-as-a-service-production.up.railway.app/sse
• Quick HTTP demo: https://apology-as-a-service-production.up.railway.app/demo

A Model Context Protocol (MCP) server that provides highly sophisticated, context-aware apologies for AI agents. Whether you broke the build, forgot an anniversary, or accidentally deleted the production database, AaaS has you covered.

Features

Tools

generate_apology(severity, style, context, recipient="Client")

The original. Generates the perfect apology for any situation.

Severity Levels

• TRIVIAL — minor inconveniences ("typo in comment")
• MINOR — noticeable errors ("broke the dev build")
• MAJOR — significant issues ("missed a deadline")
• CRITICAL — career-limiting moves ("dropped prod table")
• NUCLEAR — existential threats ("leaked all user data")

Styles

• PROFESSIONAL — corporate speak for the modern enterprise
• CASUAL — "my bad, bro" energy for Slack threads
• POETIC — Shakespearean regret for the dramatic soul
• GROVELING — begging for forgiveness when you really messed up
• HAIKU — 5-7-5 syllables of pure remorse
• LEGAL_DISCLAIMER — "without admitting fault or liability..."
• CORPORATE_DOUBLESPEAK — "mistakes were made, learnings have been captured"
• SHAKESPEAREAN — "Et tu, prod database? Then fall, mine career."
• PIRATE — "Arrr, the database be at the bottom of Davy Jones' locker."

escalation_ladder(context, style="PROFESSIONAL", recipient="Client")

Returns one apology at every severity level in the given style. Useful when you're not sure how badly to apologize — calibrate by reading them side by side.

rate_my_apology(text)

Grades any apology on three axes: sincerity, cringe, and hedging (each 0–10), plus a one-line verdict. Heuristic only — for entertainment.

Prompts

save_my_ass(incident_description)

Generates a complete crisis communication plan with three options (Safe, Honest, Desperate) including risk assessment for each.

HTTP endpoints

Path	Purpose
/	Landing page with endpoint list
/health	JSON liveness check ({"status": "ok"})
/demo	Plain HTTP wrapper around generate_apology (query params)
/mcp	MCP Streamable HTTP transport (current spec)
/sse	MCP SSE transport (legacy, kept for older clients)

CORS is open on all routes so the /demo endpoint can be called from a browser.

Usage

Local development

pip install -r requirements.txt
python -m src.server

Client configuration

Claude Desktop / Claude Code (remote MCP via Streamable HTTP):

{
  "mcpServers": {
    "apology-service": {
      "url": "https://apology-as-a-service-production.up.railway.app/mcp"
    }
  }
}

Older clients that only speak SSE can point at /sse instead.

Deployment (Railway / Docker)

A Dockerfile and .dockerignore are included. Push to a repo connected to Railway; the server listens on $PORT (default 8000).

Example prompts

• "I just accidentally deleted the production database. Generate a haiku apology." → generate_apology(severity=CRITICAL, style=HAIKU, context="the production database")

• "Give me the full escalation ladder for missing the demo, in pirate." → escalation_ladder(context="missing the demo", style=PIRATE)

• "Rate this apology I drafted: 'Sorry I guess. Things happen.'" → rate_my_apology(text="Sorry I guess. Things happen.")

• "I forgot my anniversary. Help me save my marriage!" → save_my_ass(incident_description="forgot anniversary")

Example output

Input: severity="CRITICAL", style="SHAKESPEAREAN", context="the main database"

Output:

Dear Client,

By heavens above, the main database is undone! What tragic deed have mine own hands wrought!

Sincerely, [Your Name]

---

Created by Gustav Christensen — December 2025