Getting Started

claude-mpm vs trusty-mpm — Differences & Install

The 30-Second Answer

trusty-mpm (tm) is the Rust successor to the Python claude-mpm project. Both are session orchestrators, but trusty-mpm is a rebuilt, production-grade system with:

Single binary: tm (instead of scattered Python packages)
Managed daemons: Runs trusty-memory and trusty-search as system services, not ad-hoc
Per-session git worktrees: Each session gets its own isolated branch, automatically cleaned up on decommission
MCP-native storage: Memory and search are full MCP servers, not REST wrappers

Install trusty-mpm now:

curl -sSf https://raw.githubusercontent.com/bobmatnyc/trusty-tools/main/install.sh | sh
tctl install trusty-mpm
tm --help

Then run tm in any git repo to launch a session.

For the full install walkthrough, see Install and Run trusty-mpm (tm) on Your Laptop.

What's Familiar vs. Different

Familiar: The PM/Agent/Skill Model

Both systems use the same agent orchestration framework:

	claude-mpm	trusty-mpm
PM (Project Manager)	Orchestration, delegation, handoff	Same core logic, Rust + MCP native
Agents	Specialized task runners (e.g., rust-engineer)	Same ecosystem, now deployed as JSON + embedded
Skills	Reusable slash commands (e.g., `/verify`, `/code-review`)	Same, deployed alongside agents
Output styles	Markdown/text formatting templates	Same markup, now in `~/.claude/output-styles/`

What this means: If you're familiar with claude-mpm's delegation patterns, agent selection, and output customization, trusty-mpm works the same way — it's a platform upgrade, not a concept change.

Detailed Differences Table

Aspect	claude-mpm (Python)	trusty-mpm (Rust)
Language/runtime	Python 3.x (pip install)	Rust compiled binary (`tm`)
Installation	`pip install claude-mpm`	One-liner: `curl ... \| sh` → `tctl install trusty-mpm`
Distribution	PyPI	GitHub Releases / Homebrew / `tctl`
Primary binary	`claude-mpm` (Python entry point)	`tm` / `trusty-mpm` (single binary)
Memory storage	In-process or external API	Full MCP daemon (`trusty-memory` on port 7070)
Code search	External REST API only	Full MCP daemon (`trusty-search` on port 7878)
Session management	Ad-hoc tmux windows; manual cleanup	Managed git worktrees; auto-decommission
Session isolation	Shared global state	Per-worktree branch + scoped memory palace
MCP servers	Ad-hoc wrappers	Native MCP stdio + HTTP transports
Configuration	Scattered YAML/ENV files	Unified `~/.trusty-tools/config/` + project overrides
Daemon lifecycle	Manual start/stop	System services (launchd/systemd)
Health checks	Manual queries	`tctl status`, `tm doctor`

The Three Big Structural Differences

1. Installation & Deployment

claude-mpm (Python):

pip install claude-mpm
# Plus: manually install claude-mpm-server if using memory/search APIs
# Plus: manage those server processes yourself

trusty-mpm (Rust):

curl -sSf https://raw.githubusercontent.com/bobmatnyc/trusty-tools/main/install.sh | sh
tctl install trusty-mpm
# Automatically installs + starts trusty-memory and trusty-search daemons

Impact: trusty-mpm gives you a complete platform in one install command. No manual daemon wrangling, no separate pip packages, no PATH hunting.

2. Session Lifecycle & Git Worktree Model

claude-mpm:

Creates tmux windows for each session.
Session state is ephemeral; cleanup is manual (tmux kill-session).
All sessions share the main checkout or rely on [patch.crates-io] trickery.

trusty-mpm:

Creates a dedicated git worktree per session, branched off main.
Session name is semantic (e.g., .worktrees/tm-myproject-01/) — stable across restarts.
Worktree is automatically decommissioned when the session ends (branch + directory removed).
Each session runs in an isolated branch; changes don't leak to main.

Impact: You can safely have 5 parallel sessions, each in a different branch, without branch-collision chaos or manual cleanup.

See Architecture: Memory, Sessions, Search for the full design.

3. Memory & Search as First-Class Daemons

claude-mpm:

Memory and search are optional, reached via hardcoded ports or ENV variables.
If the daemon dies, the connection silently times out or errors.
Port guessing is fragile (port 7070 may be taken → 7071 → …).

trusty-mpm:

Memory (trusty-memory) and Search (trusty-search) are always-on core services.
Ports are auto-discovered from daemon discovery files, never guessed.
Failures are explicit and detectable.
Both are full MCP servers (stdout/JSON-RPC, not ad-hoc REST).

Impact: More reliable session provisioning, better error messages, zero silent failures.

What's the Same: Agent & Skill Ecosystem

Both systems run the same agent catalog and skill library. When you tm run a task or launch a session, you get:

55+ agents deployed (rust-engineer, python-engineer, research, QA, security, etc.)
260+ skills available (verify, code-review, simplify, loop, schedule, etc.)
Output styles you can customize (how results are formatted)
Delegation patterns you're used to (agent selection, PM orchestration)

What differs: Where they live and how they're deployed.

	claude-mpm	trusty-mpm
Agent location	`~/.claude/agents/` (user config)	`~/.claude/agents/` (same) + bundled framework docs
Agent deployment	Manual YAML editing or PM updates	Provisioned by `tm load` / `tm install`
Skill location	`~/.claude/skills/` (user config)	`~/.claude/skills/` (same)
Output styles	`~/.claude/output-styles/` (user config)	Same + trusty-mpm system style

Migration Path: From claude-mpm to trusty-mpm

If you have an existing Claude Code project that was set up with claude-mpm or standalone:

Install trusty-mpm:

curl -sSf https://raw.githubusercontent.com/bobmatnyc/trusty-tools/main/install.sh | sh
tctl install trusty-mpm

Navigate to your project:
```
cd ~/your-project
```
Provision a trusty-mpm-managed workspace:
```
tm load
```
Start a new session:
```
tm
```

Your project is now managed by trusty-mpm. All future sessions will run under the trusty-mpm framework with git worktrees, managed memory, and persistent search indexes.

Architecture & Design

For a deep dive into trusty-mpm's architecture, read:

WHAT-IS-TRUSTY-MPM.md — Official identity doc. Clarifies that this is not the Python claude-mpm project. Addresses common confusion.
ARCHITECTURE-MEMORY-SESSIONS-SEARCH.md — Three-part deep dive:
1. Memory over MCP/JSON-RPC (daemon discovery, no port guessing)
2. Session↔Worktree 1:1 model (semantic naming, automatic cleanup)
3. Per-worktree search indexes (isolated, scoped to projects)
Root README — Overview of all 21 crates, including trusty-search, trusty-memory, trusty-analyze, and the agent orchestration platform.

Frequently Asked Questions

Q: Can I keep using claude-mpm if I prefer?

Yes. claude-mpm continues to work. trusty-mpm is an upgrade path, not a forced migration. That said, trusty-mpm's daemon management and session isolation make it easier to work with multiple projects in parallel.

Q: Will trusty-mpm replace claude-mpm?

Over time, yes. trusty-mpm is production-grade with all the robustness of a Rust rewrite. We recommend migrating new projects to trusty-mpm and gradually retiring the Python version.

Q: What if I have a claude-mpm project and want to keep it?

You can run both side-by-side. Each uses separate config directories (~/.trusty-tools/config/ for trusty-mpm; claude-mpm's config elsewhere). Projects can opt in to trusty-mpm by running tm load in their directory, or stay with claude-mpm — your choice.

Q: Is there a GUI for trusty-mpm?

Not yet, but trusty-console (in development) will provide a web dashboard. For now, use the CLI: tctl, tm, and tm doctor.

Q: How do I get help with trusty-mpm?

Run tm doctor to diagnose health issues.
Check Install and Run trusty-mpm for common troubleshooting.
Read the architecture docs linked above.
File an issue: https://github.com/bobmatnyc/trusty-tools/issues

Q: Does trusty-mpm work on Windows?

Currently, trusty-mpm targets macOS and Linux. Windows support via WSL2 is on the roadmap.

Next Steps

Install trusty-mpm: Install and Run trusty-mpm on Your Laptop
Deep dive: Read WHAT-IS-TRUSTY-MPM.md and ARCHITECTURE-MEMORY-SESSIONS-SEARCH.md
Explore the full platform: See the root README for trusty-search, trusty-memory, trusty-analyze, and agent orchestration