Search BrigadeStack

Search BrigadeStack pages and docs.

BrigadeStack
Menu

Comparisons

Best tools for AI agent config, MCP, and memory (2026)

There is no single best tool. Here are the 2026 options for managing MCP servers, memory, and config across coding agents, grouped by the job you are doing.

Updated November 15, 2026

Short answer: there is no single best tool, because “agent config” is several jobs: syncing MCP servers across tools, keeping memory across sessions, local evidence and code structure, and managing dotfiles. Below are the current options grouped by the job, with an honest note on each. BrigadeStack and the Escoffier Labs tools around it are on this list, and so are the tools they do not replace.

If you want to sync MCP servers across tools

  • BrigadeStack keeps one MCP catalog and merges it into each tool’s native config (Claude Code, Codex, Cursor, VS Code, OpenCode, and Grok in the project, plus Antigravity, Hermes, OpenClaw, and Kimi Code with --user-scope), dry-run by default, with a recorded projection per write. It also handles skills and memory, so this is the more complete, heavier option.
  • add-mcp installs MCP servers into 25 agent configs with native field mapping and dropped-field warnings, plus list, remove, and sync commands. Lighter: no catalog file or dry-run diff, MCP only.
  • Rulesync, agentsync, and Ruler project one source tree into many agents’ rules, skills, and MCP configs, and Rulesync and agentsync carry hooks as well. All three have dry-run modes. agentsync also reports native, lossy, and skipped fields and keeps git checkpoints you can revert, and Ruler backs files up and has a revert command. They are the closest alternatives if rules matter as much as MCP.
  • chezmoi is a general dotfile manager. You can template MCP configs with it and it shows a per-file diff before it applies, but it does not understand each tool’s shape, so there is no per-server merge or per-tool plan.

If you want memory that persists across sessions

  • mem0 is a memory layer with an SDK, a REST API, a hosted MCP server, and plugins for Claude Code and Codex. Strongest when you are building a product that remembers its users, and usable from coding agents too. Memories live in your Mem0 account or a server you host.
  • Letta (formerly MemGPT) is now an open agent harness, Letta Code, with a git-backed memory filesystem and background “dreaming”. It replaces your coding CLI more than it adds memory to it. The older memory-block server is legacy.
  • agentmemory gives coding-agent CLIs persistent memory and handoffs across tools through an MCP server and hooks, focused on memory.
  • BrigadeStack keeps memory as markdown cards in your repo with a slim per-session index, alongside MCP and tools. In the 0.32.0 beta, memory care also fixes and archives cards on its own, with undo.
  • Native harness memory (each tool’s own memory) is the zero-setup option. Claude Code writes auto memory on its own, and Codex memories are opt-in. Each store is per tool, and shared instruction files such as AGENTS.md are the part that crosses agents.

If you want local structure and evidence (built into BrigadeStack)

  • Code intelligence is BrigadeStack’s built-in local code graph: callers, callees, impact for agents. See local code graph for AI coding agents.
  • Evidence memory is BrigadeStack’s built-in local evidence ledger over sessions and receipts, with crawl planning and receipt export. See local evidence ledger for agent work.
  • Agent Pantry, Agent Notify, Skillet, and Memory Doctor cover auth sync, notifications, skills, and memory maintenance. BrigadeStack built-ins and sidecars stay on brigadestack.com, and standalone kitchen tools use *.escoffierlabs.dev.

If you want to manage dotfiles in general

  • chezmoi and GNU Stow are the standard answers. They are not agent-aware, but if you mainly want your shell and editor configs versioned, they are the right tool for that, and they stay a dotfile layer.

If you want to run one task across several agents

  • BrigadeStack run takes one plain-language task, derives what it needs with a deterministic router (brigadestack route "<task>" prints the plan first), and executes the required stages as dependency-ordered parallel waves across the CLIs you already pay for, then synthesizes the results into a receipt. The plan comes from task signals you can inspect and override (--route-signal +auth-surface), not from a model inventing its own steps.

How to choose

Your jobReach for
Same MCP servers in every coding agentBrigadeStack for a reviewed catalog, or add-mcp for a quick install across many clients
Rules, skills, and MCP projected from one treeRulesync, agentsync, Ruler, or BrigadeStack
Memory inside a product you are buildingmem0
Memory across the coding-agent CLIs you runBrigadeStack or agentmemory (handoffs)
One reviewed source for MCP + skills + memory, with receiptsBrigadeStack
Local code graph / impact for agentsBrigadeStack code intelligence
Local searchable agent work evidenceBrigadeStack evidence memory
Run one task across several agent CLIsBrigadeStack (brigadestack route derives signals, then runs dependency-ordered waves)
Just version my dotfileschezmoi or Stow

The honest summary: if your problem is only one of these, a focused tool is lighter. BrigadeStack is for the case where you have several coding agents and want their MCP servers, tools, and memory to come from one reviewed source, with a plan before config changes and a record after.

Questions

Is BrigadeStack the best memory tool? No, and it does not try to be. mem0 is stronger if your problem is programmable memory inside an app. BrigadeStack is the best fit when you run multiple coding agents and want their config, skills, and memory unified and reviewable on disk.

What is the lightest way to just share MCP servers? add-mcp for a quick install across many clients, or BrigadeStack if you want a catalog and a dry-run diff. See syncing MCP servers across coding agents, or the narrower Claude Code and Codex page if that is your exact setup.