Home

MCE — MCP Code Execution

APIs were designed for developers. MCE recompiles them for AI.

MCE is a production-grade MCP (Model Context Protocol) server that converts Swagger/OpenAPI specifications into LLM-native Python functions and executes them inside isolated Docker containers. Instead of flooding LLM context with hundreds of endpoint-specific tools, MCE exposes exactly 4 meta-tools + 1 prompt — letting the model discover, inspect, write, and reuse code against any number of APIs.

---

Navigation

Section	Description
Architecture Overview	How MCE is structured end-to-end
Getting Started	Installation, configuration, and first run
MCP Tools Reference	All 4 tools + 1 prompt, with examples
Configuration Reference	Environment variables and swagger config
Security Model	Defense-in-depth layers and credential isolation
Compiler Pipeline	How Swagger specs become Python functions
SIMD Pattern	Writing reusable, cacheable code
Contributing	Setup, workflow, coding standards, PR checklist

---

Why MCE?

When integrating REST APIs with LLMs via MCP, the naive approach — one tool per endpoint — creates compounding problems:

Problem	Impact
Context window bloat	A 200-endpoint API burns hundreds of tokens per call just describing tools
Tool processing limits	MCP clients cap tool counts; large APIs hit the limit and fail silently
Insecure code execution	Running LLM-generated code on the host is dangerous
Bloated responses	Raw API responses include metadata, nulls, and deprecated fields the model never needs
Integration friction	Developers spend days writing auth wrappers and prompt scaffolding for every new API

MCE eliminates all five problems with a single pattern: compile once, discover–inspect–execute–reuse.

---

Architecture Overview

graph TB
    subgraph MCE["MCE MCP Server"]
        Compiler["Compiler\n(setup)"]
        Runtime["Runtime\n(serve)"]
        Executor["Code Executor\n(Docker SDK)"]

        subgraph Core["Core Services"]
            SwaggerParser
            FunctionRegistry
            CacheStore
            SecurityGuard
            CredentialVault
        end

        subgraph MCPTools["4 MCP Tools (exposed to LLM)"]
            list_servers
            get_functions
            execute_code
            run_cached_code
        end

        Compiler --> Core
        Runtime --> Core
        Executor --> Core
    end

    MCE --> Swagger["Swagger Sources"]
    MCE --> Docker["python:3.13-slim\nDocker Container"]

Loading

Subsystems at a Glance

Subsystem	Responsibility
Compiler	Parses Swagger/OpenAPI specs, generates typed Python functions via Jinja2, writes manifests
Runtime	Loads compiled manifests at startup, serves function metadata to MCP tools
Executor	Validates code (size → AST guard → lint), runs it in a Docker sandbox, returns results
Cache	Async SQLite store keyed by code hash; enables the SIMD reuse pattern
Security	AST guard, credential vault, domain allowlist, read-only enforcement

LLM Workflow

graph LR
    A["list_servers\n(discover)"] --> B["get_functions\n(inspect)"] --> C["execute_code\n(execute)"] --> D["run_cached_code\n(reuse)"]

Loading

---

Quick Example

# 1. Discover available APIs
list_servers()
# → { "servers": [{ "name": "weather", "functions": ["get_current_weather", ...] }] }

# 2. Inspect the function signature
get_functions([{"server_name": "weather", "function_name": "get_current_weather"}])
# → { "functions": [{ "parameters": [...], "import_statement": "from weather.functions import get_current_weather", ... }] }

# 3. Write and execute code
execute_code("""
from weather.functions import get_current_weather

city = "London"    # top-level variable — the only thing that changes per request

def main():
    return get_current_weather(city=city, units="metric")

result = main()
""", description="get weather by city")
# → { "success": true, "data": { "temperature": 15.2, "condition": "Cloudy" }, "cache_id": "abc123" }

# 4. Reuse for a different city — no rewriting code
run_cached_code("abc123", params={"city": "Paris"})
# → { "success": true, "data": { "temperature": 18.5, "condition": "Sunny" }, "cache_id": "def456" }

---

Project Status

Item	Status
Version	0.1.0 (Beta)
Python	3.13+
License	MIT
Test coverage gate	≥ 90%
CI	GitHub Actions

---

Continue to Getting Started →