callite 1.0.5

Slim Redis RPC implementation

Callite

Callite is a lightweight Remote Procedure Call (RPC) library over Redis, designed for communication between components of a distributed system. It uses Redis Streams for request transport and Redis Pub/Sub for response delivery, with pickle serialization.

Installation

pip install callite

To use the MCP bridge (for AI agent integration):

pip install callite[mcp]

Prerequisites

A running Redis instance is required. You can start one locally:

docker run -d -p 6379:6379 redis:alpine

Or use the included Docker Compose setup for a full development environment:

docker-compose up

Usage

Request/Response (register + execute)

The most common pattern: the client sends a request and waits for a response.

Server:

from callite.server import RPCServer

server = RPCServer("redis://localhost:6379/0", "my_service")

@server.register
def healthcheck():
    return "OK"

@server.register
def add(a, b):
    return a + b

server.run_forever()

Client:

from callite.client import RPCClient

client = RPCClient("redis://localhost:6379/0", "my_service")

# Positional arguments
result = client.execute("add", 1, 2)
print(result)  # 3

# Keyword arguments
result = client.execute("add", a=10, b=20)
print(result)  # 30

# No arguments
status = client.execute("healthcheck")
print(status)  # "OK"

# Close when done
client.close()

execute() blocks until the server responds or the timeout is reached (default: 30 seconds). If the server-side handler raises an exception, execute() re-raises it on the client side.

Fire-and-Forget (subscribe + publish)

For one-way messages where the client does not need a response.

Server:

from callite.server import RPCServer

server = RPCServer("redis://localhost:6379/0", "my_service")

@server.subscribe
def log(message):
    print(f"Received: {message}")

server.run_forever()

Client:

from callite.client import RPCClient

client = RPCClient("redis://localhost:6379/0", "my_service")

# publish() returns immediately without waiting for a response
client.publish("log", "Something happened")

Combining Both Patterns

A single service can use both register and subscribe:

from callite.server import RPCServer

server = RPCServer("redis://localhost:6379/0", "my_service")

@server.register
def add(a, b):
    return a + b

@server.subscribe
def log(message):
    print(message)

server.run_forever()

from callite.client import RPCClient

client = RPCClient("redis://localhost:6379/0", "my_service")

# Request/response
result = client.execute("add", 3, 4)

# Fire-and-forget
client.publish("log", f"Result was {result}")

MCP Integration (AI Agent Integration)

Callite includes three MCP (Model Context Protocol) components for bridging AI agents to RPC services. All three are available from callite.mcp:

from callite.mcp import MCPBridge, MCPHTTPProxy, MCPProxy

Defining Tools and Prompts

register_tool works like register but extracts type hints and docstrings to generate rich metadata for MCP tool discovery:

from callite.server import RPCServer

server = RPCServer("redis://localhost:6379/0", "data_service")

@server.register_tool(description="Add two numbers together")
def add(a: int, b: int) -> int:
    """Add two numbers.

    Args:
        a: The first number.
        b: The second number.
    """
    return a + b

server.run_forever()

register_prompt registers prompt templates that AI agents can discover and invoke:

@server.register_prompt(description="Analyze a dataset")
def analyze(data: str, focus: str = "general") -> str:
    return f"Please analyze the following data with focus on {focus}:\n{data}"

MCPBridge

A multi-service MCP gateway that discovers callite services via their __describe__ endpoint and registers all tools, prompts, and resources as MCP primitives. Tool names are prefixed with the service name (e.g. data_service_add).

From the command line:

mcp-callite-bridge --redis redis://localhost:6379/0 --services data_service

Or programmatically:

from callite.mcp import MCPBridge

bridge = MCPBridge("redis://localhost:6379/0", ["data_service", "auth_service"])
bridge.run()  # stdio transport by default

CLI options:

Flag	Description	Default
--redis	Redis connection URL	redis://localhost:6379/0
--services	Comma-separated service names	(required)
--transport	stdio or streamable-http	stdio
--name	MCP server display name	callite-bridge
--timeout	RPC execution timeout (seconds)	30
--queue-prefix	Redis key prefix	/callite

MCPHTTPProxy

An HTTP server that exposes a single callite service over Streamable HTTP. Unlike MCPBridge, tool names are not prefixed with the service name, making it suitable for single-service deployments.

from callite.client import RPCClient
from callite.mcp import MCPHTTPProxy

client = RPCClient("redis://localhost:6379/0", "data_service")
proxy = MCPHTTPProxy(client, host="0.0.0.0", port=8080)
proxy.run()  # serves at /mcp

MCPProxy

A client-side proxy that manages an external MCP server subprocess (e.g. uvx mcp-server-sqlite) over stdio and exposes its tools through a synchronous Python API. Use register_proxy on an RPCServer to re-publish external MCP tools as callite RPC methods.

from callite.mcp import MCPProxy
from callite.server import RPCServer

server = RPCServer("redis://localhost:6379/0", "my_service")

proxy = MCPProxy("uvx", ["mcp-server-sqlite", "--db-path", "test.db"])
server.register_proxy(proxy, prefix="sqlite")

server.run_forever()

The proxy auto-reconnects if the subprocess crashes, caches the tool list, and is thread-safe.

Configuration

Environment Variables

Variable	Description	Default
LOG_LEVEL	Logging verbosity (DEBUG, INFO, ERROR, etc.)	ERROR (server), INFO (client)
EXECUTION_TIMEOUT	Client-side timeout in seconds for execute()	30
REDIS_URL	Redis URL (used by the MCP bridge CLI)	redis://localhost:6379/0

Constructor Options

RPCServer:

RPCServer(
    conn_url="redis://localhost:6379/0",
    service="my_service",
    queue_prefix="/callite",       # Redis key prefix
    xread_groupname="generic",     # Consumer group name
)

RPCClient:

RPCClient(
    conn_url="redis://localhost:6379/0",
    service="my_service",
    execution_timeout=30,          # Timeout in seconds
    queue_prefix="/callite",       # Redis key prefix
)

Docker Development

The included docker-compose.yml starts Redis, a sample server, and a sample client:

docker-compose up

This runs the example main.py (server) and healthcheck.py (client stress test with 100 concurrent threads).

License

Proprietary