TECHNICAL ARCHITECTURE

The execution layer between AI models and enterprise systems.

LitEngine implements the open Model Context Protocol (MCP) to provide AI assistants with deterministic, schema-typed access to business databases, statutory registries, and operational workflows.

OPEN STANDARD

Built natively on the Model Context Protocol

The Model Context Protocol (MCP) is an open specification that standardizes how AI applications discover and invoke tools, query context, and interact with external systems.

Standardized JSON-RPC 2.0:

Every tool invocation, schema handshake, and progress event follows standard JSON-RPC 2.0 messages over stdio or Server-Sent Events (SSE).

Zero Vendor Lock-in:

Because LitEngine uses standard MCP, your AI workflows are not locked to a specific LLM vendor. Switch between Claude 3.5 Sonnet, Claude Opus, or custom open models with zero protocol rewrites.

mcp_handshake.jsonJSON-RPC 2.0
// 1. Client initializes session
--> {"jsonrpc":"2.0","method":"initialize",
     "params":{"protocolVersion":"2024-11-05"}}

// 2. LitEngine returns available capabilities
<-- {"jsonrpc":"2.0","result":{
       "capabilities":{
         "tools":{"listChanged":true},
         "resources":{}
       },
       "serverInfo":{"name":"litengine-mcp","version":"2.4.0"}
     }}

// 3. AI invokes balance sheet tool
--> {"jsonrpc":"2.0","method":"tools/call",
     "params":{
       "name":"extract_xbrl_financials",
       "arguments":{"companyNumber":"11849201"}
     }}
CONNECTIVITY

Seamless connection with leading AI clients

How your reasoning client discovers and securely interacts with LitEngine's operational tools.

DESKTOP WORKFLOWS

Claude Desktop

Connect Anthropic's Claude Desktop directly to LitEngine by adding one entry to your claude_desktop_config.json. Claude immediately inherits all statutory research and CRM capabilities.

ENGINEERING AGENTS

Cursor & Developer IDEs

Developers building inside Cursor can query live company financials, test outreach hooks, and inspect CRM databases without leaving their coding workflow.

ENTERPRISE AGENTS

Custom SDK Runtimes

Integrate with your own Python, TypeScript, or Go agent services via the official Model Context Protocol SDKs or standard SSE streaming transports.

SPECIALIST TOOLS

Deterministic data access and execution

Exposing grounded business data to AI assistants through typed, validated schemas.

Companies House & Statutory Registries

LitEngine maintains direct programmatic pipelines into the UK Companies House REST API and bulk snapshot archives. When an AI client asks about a firm, it receives verified registration status, active officers, filing histories, and PSC data.

> query_company("Halsbury Homes") → CRN 11849201 [Verified Active]

Automated XBRL Balance Sheet Extraction

Unstructured iXBRL and statutory PDF accounts are converted into certified balance sheet metrics: Current Assets, Creditors due within one year, Net Assets, Working Capital, and calculated liquidity ratios.

> extract_financials() → Net Assets: -£2.14m | Ratio: 0.68
ENTERPRISE GOVERNANCE

Secure authentication and human-in-the-loop review

Autonomy without safety is a liability. LitEngine enforces strict security boundaries between AI reasoning and real-world system actions.

Scoped API Credentials

Grant read-only registry lookup access without permitting write operations to your CRM or email dispatchers.

Human Approval Queue

High-stakes actions—such as sending outreach or modifying deal status—are routed into an operator queue for human verification.

Complete Audit Logs

Every model query, tool request, and execution result is timestamped and recorded in full for compliance and governance review.

Governance Checklist

Deterministic tool argument validation
Zero training on client data
Tone and compliance enforcement guardrails
Granular rate-limiting per MCP client token
Multi-tenant workspace isolation
ORCHESTRATION

Graph-based workflow execution engine

Orchestrate complex business logic across parallel branches, conditional guards, and automated retries.

DAG EXECUTION

Directed Acyclic Graphs

Execute dependency graphs where tasks run concurrently the moment their upstream prerequisites resolve, minimizing latency.

CONDITIONAL BRANCHING

Dynamic Routing

Route execution paths based on evaluated business conditions, such as balance sheet liquidity ratios or sector SIC categories.

RESILIENCE

Automatic Retries & Error Handling

Built-in exponential backoff for external API rate limits and non-blocking failure boundaries ensuring continuous operation.

EXECUTION AT SCALE

Ready to deploy LitEngine infrastructure?

Connect Claude Desktop in minutes or schedule an engineering consultation for custom enterprise deployments.

Compatible with Claude Desktop, Cursor, LangChain & standard MCP clients.