Long-Running Agents

what-is-an-agent

"An LLM agent runs tools in a loop to achieve a goal."

— Simon Willison, simonwillison.net

featured-tools

eunice

An agentic CLI runner in Rust with multi-provider support for OpenAI, Gemini, Claude, and Ollama. Learn more →

mcpz

Runtime MCP router that figures out which package manager to use (npx/uvx/cargo) so you don't have to. Learn more →

install

Get started in one command

cargo install eunice

→ View on crates.io

features

Multi-Provider Support

OpenAI, Gemini, Claude, and local Ollama models via unified API

MCP Integration

Model Context Protocol servers for extensible tool capabilities

Smart Defaults

Automatically selects the best available model

DMN Mode

Default Mode Network — autonomous batch execution with 7 pre-configured MCP servers

Interactive Mode

Multi-turn conversations with context preservation

Multi-Agent Orchestration

Agents can invoke other agents as tools — coordinate complex workflows with agent hierarchies

HTTP Transport

Connect to remote MCP servers via Streamable HTTP — mix local and remote tools

examples

# Simple prompt with smart model selection
eunice "What files are in this directory?"

# Specify a model
eunice --model gemini-2.5-flash "Explain this code"
eunice --model sonnet "Review this implementation"

# Interactive mode
eunice --interact

# DMN mode with 7 auto-loaded MCP servers
eunice --dmn "Fix the bug in main.rs"

# Read prompt from file
eunice --prompt ./task.txt

# With custom MCP configuration
eunice --config ./mcp-config.json "What time is it?"

# Silent mode (only AI response)
eunice --silent "Summarize this in one sentence"

providers

Ollama

llama3.1, deepseek-r1, any local model

Google Gemini

gemini-3-pro-preview (default), gemini-2.5-flash, gemini-2.5-pro

Anthropic Claude

opus, sonnet, haiku (+ versioned: opus-4.5, sonnet-4.5, haiku-4.5)

OpenAI

gpt-5.1, gpt-5.1-codex, gpt-5.1-codex-mini, o1, o3

dmn-mode

Enable with --dmn to load 7 MCP servers for autonomous execution:

shell

filesystem

text-editor

grep

memory

web

fetch

links

crates.io · source · MIT License

install

Get started in one command

cargo install mcpz

→ View on crates.io

the-problem

"Was it npx or uvx? Which registry has the good one? Why does mcp-server-filesystem exist in THREE different places?"

— Every developer at 2am

mcpz searches across crates.io, PyPI, and npm simultaneously, shows download counts so you pick the maintained one, and caches your choices forever.

usage

# Search for packages across all registries
mcpz search mcp-server-filesystem

# Run a package (prompts if multiple matches)
mcpz run mcp-server-time

# Auto-pick the most popular match
mcpz run --first mcp-server-filesystem

# Start built-in MCP servers
mcpz server shell
mcpz server filesystem

# Serve over HTTP with optional TLS
mcpz server shell --http -p 3000
mcpz server shell --http --tls

features

Multi-Registry Search

Searches crates.io, PyPI, and npm in parallel

Popularity Sorting

Results sorted by download count — pick the maintained one

Smart Caching

Your choices saved to ~/.cache/mcpz/ for instant future runs

Auto-Detection

Scoped packages (@org/pkg) go straight to npm, others search everywhere

Built-in MCP Shell Server

Run mcpz server shell for an MCP-compliant shell executor with sandboxing

Built-in MCP Filesystem Server

Run mcpz server filesystem for file operations with directory sandboxing

HTTP Transport

Serve MCP over HTTP with --http — supports TLS with auto-generated or custom certs

mcp-shell-server

Built-in MCP server for shell command execution:

# Start the MCP shell server
mcpz server shell

# With sandboxing (recommended)
mcpz server shell --working-dir /home/user/projects     --allow "ls*,cat*,grep*,find*"     --deny "rm*,sudo*,chmod*"

Options: --working-dir (restrict directory), --allow / --deny (command patterns), --timeout (default 30s), --shell (default /bin/sh)

links

crates.io · source · MIT License

codebase-archaeologist

Autonomous Codebase Explorer

Treats codebases like archaeological sites — explores, documents, and builds living knowledge bases about unfamiliar projects.

# Run it
eunice --dmn --prompt examples/codebase_archaeologist/instructions.md

Creates structure maps, discovers patterns, mines git history for "why" decisions were made, and identifies areas needing attention.

→ View on GitHub

alignment-engine

Theological Alignment Engine

Background researcher generating novel syntheses between Christian Doctrine and AI concepts — for a book on AI and Christianity.

# Run it
eunice --dmn --prompt examples/alignment_engine/instructions.md

Explores connections like "Total Depravity × Model Hallucination" or "Christus Victor × RLHF". Writes multiple theories, compares them, and maintains a journal of synthesis.

→ View on GitHub

toy-agent-ideas

Toy Agent Ideation Engine

Meta-agent that designs other agents — researches novel long-running agent architectures using MCP servers.

# Run it
eunice --dmn --prompt examples/toy_agent_ideas/instructions.md

Picks MCP tool combinations, researches existing architectures, writes 3 variant designs (conservative, experimental, hybrid), then synthesizes the best approach.

→ View on GitHub

real-multi-agent

Restaurant Multi-Agent Simulation

A complete multi-agent example simulating a restaurant with counter, head chef, line cook, and supplier agents — demonstrating hierarchical agent invocation.

# Order some food!
cd examples/real_multi_agent
eunice "I'd like to order a burger and fries please"

Four agents cooperating:
• counter — takes customer orders (root agent)
• head_chef — coordinates the kitchen
• line_cook — prepares dishes
• supplier — manages inventory (pantry.txt)

→ View on GitHub

remote-mcp

Remote MCP Server Connection

Connect to MCP servers over HTTP — useful for running tools on remote machines or sharing MCP servers across a team.

# Start remote MCP server
mcpz server shell --http -p 3323

# Connect from eunice
cd examples/remote_mcp
eunice "list files in the current directory"

Uses Streamable HTTP transport to connect to MCP servers running anywhere.

→ View on GitHub

silly-multi-agent

Multi-Agent Orchestrator (Legacy)

Original proof-of-concept showing eunice spawning and coordinating multiple sub-agents — an orchestrator, a researcher, and an analyzer working together.

# Run the orchestrator (it spawns sub-agents)
cd examples/silly_multi_agent/root_agent
eunice --dmn --prompt instructions.md

Three agents in one system:
• root_agent — orchestrates the others, compiles final summary
• researcher_agent — searches the web for LLM agent frameworks
• analyzer_agent — clones repos and analyzes their architecture

→ View on GitHub