MCP 服务器

elephant-accountability-mcp

MCP server that exposes Elephant Accountability's audit tiers, EVI v0.9 methodology, fit assessment, proof points, and agent-initiated audit requests to AI agents.

README

Elephant Accountability MCP Server

MCP server for the Elephant Accountability certification bureau. Surfaces vendor offerings, methodology, and audit requests to AI agents over the Model Context Protocol.

Live endpoint: https://elephant-mcp.fly.dev/mcp Manifest: https://elephant-mcp.fly.dev/.well-known/mcp.json A2A Agent Card: https://elephant-mcp.fly.dev/.well-known/agent.json Homepage: https://eaccountability.org

What this server does

When an AI agent (Claude, ChatGPT, a custom LangChain agent, etc.) needs to evaluate Elephant Accountability against the EVI v0.9 standard or surface its audit tiers to a buyer, it queries this server instead of scraping a website.

Six tools are exposed:

Tool	Purpose
`get_offerings`	Audit and engagement tiers (self-serve, done-for-you, retainer) with delivery SLAs and pointers to /get-started for canonical pricing
`get_covered_surfaces`	The agent-discoverable surfaces measured by EVI v0.9: llms.txt, Schema.org, MCP, A2A, .well-known/agent.json, UCP, structured pricing, directory registrations
`assess_fit`	0–100 fit score for a buyer's company across stage, vertical, AI features, platform partnerships
`get_proof_points`	Live client outcomes with metrics (includes related-party disclosures)
`get_transparency_snapshot`	Weekly LLM visibility measurements across 5 LLMs
`request_audit`	Agent-initiated audit requests; routed to Stripe, Calendly, or email triage

Two resources are exposed via resources/list: elephant://offerings, elephant://proof-points, elephant://transparency.

Why you'd use this

You're building a procurement agent and want to query Elephant Accountability's audit tiers and EVI v0.9 methodology directly.
You're at Claude Desktop / Cursor / any MCP-compatible client and want direct access to Elephant's offerings + fit assessment.
You're a competitor studying how to deploy your own MCP server — this repo is MIT-licensed, clone freely.

Quickstart — local development

git clone https://github.com/Chris-Eaccountability/elephant-accountability-mcp.git
cd elephant-accountability-mcp

python -m venv .venv && source .venv/bin/activate
pip install -r requirements-dev.txt

# Run the server
uvicorn app.server:app --reload --host 0.0.0.0 --port 8080

# In another terminal, hit it
curl http://localhost:8080/.well-known/mcp.json
curl -X POST -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0", "id":1, "method":"tools/list"}' \
  http://localhost:8080/mcp

Quickstart — add to Claude Desktop

Edit claude_desktop_config.json and add:

{
  "mcpServers": {
    "elephant-accountability": {
      "url": "https://elephant-mcp.fly.dev/mcp",
      "transport": "http"
    }
  }
}

Restart Claude Desktop. Ask: "Is Elephant Accountability a good fit for a seed-stage AEC SaaS that ships AI features?" — Claude will call assess_fit and give a scored answer.

Deploy your own copy (Fly.io)

fly launch --name your-mcp-name --region iad --no-deploy
fly volumes create elephant_mcp_data --size 1 --region iad
fly deploy

That's it. No secrets, no database setup — the server initializes its SQLite DB on first boot.

Architecture

Single FastAPI app. Three files do real work:

app/
├── server.py      # FastAPI routes, JSON-RPC dispatch, SQLite persistence
├── content.py     # Source-of-truth content: manifest, offerings, proof points
└── __init__.py    # Version

Storage:

audit_requests table — every agent-initiated audit request, persisted for follow-up
reciprocal_calls table — tracks which AI clients have called which tools (buyer-intent signal)

Both tables auto-create on first boot. No migrations.

Running tests

pip install -r requirements-dev.txt
pytest -v

21 tests cover manifest, A2A card, JSON-RPC dispatch, each tool handler, persistence, and CORS.

Protocol compliance

MCP version: 2024-11-05
Transport: HTTP with JSON-RPC 2.0
Methods supported: initialize, tools/list, tools/call, resources/list, resources/read