Add Model Context Protocol (MCP) server for Allure Docker Service #291

Summary

Add a Model Context Protocol (MCP) server that lets LLM agents (Claude Desktop, Claude Code, and any MCP-compatible host) drive a running allure-docker-service instance through natural-language tool calls.

Motivation

allure-docker-service already exposes a complete REST API for managing projects, sending results, and generating/exporting Allure reports. Exposing that same surface as MCP tools means an AI assistant can, for example:

"Create a project regression-suite, send the results from ./allure-results, generate a report, and give me the report URL."

…without the user writing any curl/HTTP glue. This makes the service first-class in agentic CI/test-triage workflows.

Proposed scope

A self-contained allure-docker-mcp/ package that is a client of the existing API (it does not modify or replace the service):

• One MCP tool per REST endpoint: get_version, get_config, list_projects, get_project, search_projects, create_project, delete_project, send_results, send_results_from_directory (convenience), generate_report, get_latest_report, clean_results, clean_history, export_report, export_emailable_report, render_emailable_report.
• Transparent support for the service's optional security mode (SECURITY_ENABLED=1): JWT cookie login + X-CSRF-TOKEN on mutating requests.
• Configuration via environment variables (ALLURE_ENDPOINT, ALLURE_USERNAME, ALLURE_PASSWORD, …).
• stdio and streamable-http transports, plus a Dockerfile.
• Tests (unit + mocked API) and docs.

Acceptance criteria

• Each tool maps faithfully to its REST endpoint (method, params, body).
• Security mode (login + CSRF) works end-to-end.
• Unit tests pass and the package installs cleanly (pip install .).
• README documents configuration and client setup.

Implementation tracked in the linked PR.