Prismatic MCP

Overview

Prismatic MCP implements the Model Context Protocol, providing a standardized interface for AI models (Claude, GPT, local Ollama models) to interact with the Prismatic Platform. It exposes 27 tools covering OSINT queries, security operations, entity management, and platform administration through the MCP specification.

The Model Context Protocol is an open standard that enables AI assistants to interact with external tools and data sources in a structured, validated, and auditable manner. Prismatic MCP transforms the platform's capabilities into MCP-compliant tool definitions, allowing any MCP-capable AI client to query OSINT sources, run security scans, manage entities, and monitor platform health through a single standardized interface.

Architecture

Prismatic.MCP.Application
└── Prismatic.MCP.Supervisor (:one_for_one)
    ├── Prismatic.MCP.Server (GenServer)
    │   └── MCP protocol handler (JSON-RPC 2.0)
    ├── Prismatic.MCP.Transport (GenServer)
    │   ├── :stdio (local Claude/GPT)
    │   └── :sse (remote clients via Plug)
    ├── Prismatic.MCP.Router (GenServer)
    │   └── Pattern-matched tool dispatch
    ├── Prismatic.MCP.ToolRegistry (GenServer)
    │   └── ETS: :mcp_tools (27 tool definitions, hot reload)
    ├── Prismatic.MCP.Auth (GenServer)
    │   └── RBAC with JWT token validation
    └── Prismatic.MCP.AuditLog (GenServer)
        └── Immutable audit trail for all invocations

MCP Client (Claude/GPT) --> Transport Layer --> Router --> Tool Handler --> Platform API --> Result
         |                       |                |            |                |             |
    JSON-RPC 2.0            stdio/SSE       Tool lookup    Validation    Prismatic.*     Formatted
    Messages               transport         + dispatch    + auth       facade call      response

Key Modules

Available Tools (27)

OSINT Tools (8)

Security Tools (6)

Storage Tools (5)

Analysis Tools (4) and Admin Tools (4)

Configuration

config :prismatic_mcp, :server,
  transport: :stdio,           # or :sse for remote
  port: 3000,                  # SSE port (if applicable)
  tools_module: PrismaticMcp.Tools,
  auth_module: PrismaticMcp.Auth,
  rate_limit: %{
    requests_per_minute: 60,
    burst: 10
  },
  audit_log: true

config :prismatic_mcp, :auth,
  roles: [:reader, :analyst, :admin],
  default_role: :reader

API Reference

# Direct tool invocation from Elixir
{:ok, result} = PrismaticMcp.invoke("search_entity", %{
  query: "example.com",
  sources: ["shodan", "censys"],
  max_results: 20
})

# Batch tool invocation
{:ok, results} = PrismaticMcp.batch_invoke([
  {"search_entity", %{query: "example.com"}},
  {"check_rating", %{target: "example.com"}},
  {"dns_enumeration", %{domain: "example.com"}}
])

# Tool discovery (MCP clients use tools/list endpoint)
{:ok, tools} = PrismaticMcp.list_tools()

Security: Role-Based Access Control

Testing

# Run all MCP tests
cd apps/prismatic_mcp && mix test

# Run with coverage
mix test --cover

# Run tool handler tests
mix test test/prismatic_mcp/tools_test.exs

# Run auth and RBAC tests
mix test test/prismatic_mcp/auth_test.exs

Testing covers tool invocation with parameter validation, RBAC permission enforcement, audit log completeness, rate limiting behavior, and transport layer protocol compliance. Integration tests verify end-to-end tool execution against platform facade APIs.

Integration Points

NABLA Compliance

Performance

Related Resources

• Prismatic Ollama -- Local AI inference
• Prismatic Claude -- Claude AI integration
• Prismatic API -- REST API gateway
• AIAD Standard -- MCP tool definitions follow AIAD specifications
• Cross-Domain Flexibility -- 27 tools spanning 5 domains
• Quality Gates -- Validation, rate limiting, and audit logging gates

Connect & Contribute

Created by Tomáš Korcak (korczis) | Open Source under GHL

• GitHub | GitLab | LinkedIn | Contact
• Developer Portal | Architecture | Meet the Creator