Interacts | MCP Server for Fluid API | Fluid Attacks Help

Introduction

Interacts is a Model Context Protocol (MCP) server that allows AI systems to connect and work with the Fluid Attacks platform over the Internet. To learn how to configure it or how to use it with our AI Agent, read the articles "Install the MCP for AI tool integration," "Capabilities and use cases of the MCP," and "Ask the AI Agent," respectively.

Core architecture

MCP Server (mcp/)

Exposes the Fluid Attacks platform capabilities as MCP tools:

  1. Vulnerability management (query, filter, analyze)
    1. fetch_group_weaknesses
    2. fetch_weakness_vulnerabilities
    3. get_vulnerability_details
    4. get_group_weaknesses_report
  2. Analytics and reporting (risk metrics, trends)
    1. get_organization_analytics
    2. get_group_analytics
  3. Asset discovery (Git roots, IP roots, URL roots)
    1. get_group_git_roots
    2. get_group_ip_roots
    3. get_group_url_roots
    4. fetch_group_root_vulnerabilities
  4. DevSecOps integration (Forces agent, CI/CD)
    1. get_devsecops_agent_executions
    2. get_unsolved_events
  5. Organization and group management
    1. get_organization_groups
    2. get_organization_groups_information
    3. get_group_information
    4. get_group_remediation_information
    5. get_group_weaknesses_overview
  6. Knowledge base search (Kendra integration)
    1. search_related_articles
    2. get_articles
  7. Prompts for common workflows (scanners, reports, GitHub CI)
    1. run_sca_scanner
    2. run_sast_scanner
    3. run_sca_and_sast_scanners
  8. GraphQL queries to the Fluid Attacks API
    1. query
    2. describe_graphql_type

FastAPI application (app/)

REST API endpoints:

  1. /streaming - Streaming conversation endpoint
  2. /conversation_context_request - Retrieve conversation history
  3. /reset_conversation_context - Clear conversation history
  4. /_used_tools - Get tools used in conversations (Used for evaluation)
  5. /health - Health check

MCP server mounted at /mcp

AI Agent system (agent/)

  1. Streaming conversation handling
  2. Memory management with chat caching (Valkey/Glide)
  3. Conversation resumption
  4. Guardrails for safe interactions
  5. Multi-model support (Claude Sonnet 4.5, Sonnet 4, Haiku)

Key features

  • Read-only access: queries platform data; no modifications
  • Multi-language support: responds in the user's language
  • Conversation memory: maintains context across interactions
  • Streaming responses: real-time streaming
  • Observability: Logfire tracing and Bugsnag error tracking
  • Security: token-based authentication with user validation

Technology stack

  • FastAPI for the web framework
  • http-mcp for MCP server implementation
  • Pydantic AI with Bedrock for LLM integration
  • Rath for GraphQL client operations
  • Valkey/Glide for conversation caching
  • LangSmith for tracing and monitoring