agents

One config to rule them all.
Practical standard layer for multi-LLM development.

Quick Start · Integrations · Commands · Plugins · FAQ

The Problem

Instructions converged on AGENTS.md and skills converged on SKILL.md. MCP server configuration did not. Every tool keeps its own file, in its own format, in its own place:

Add a server by hand and you write it 18 times, then watch the copies drift apart.

agents keeps one source of truth in .agents/ and materializes it for every tool you enable.

Quick Start

# 1. Install
npm install -g @agents-dev/cli

# 2. Interactive setup — picks integrations, adds MCP servers, syncs everything
agents start

# 3. Re-sync whenever config changes
agents sync

.agents/agents.json is now the single source of truth. agents sync --check exits 2 when a tool config has drifted, so it works as a CI gate.

Supported Integrations

Eighteen tools. "Verified" means the tool's own CLI was run against the file this project writes; the rest follow vendor documentation and are worth reporting on if something is off.

Skills column. bridge means a symlink from the tool's directory to .agents/skills (a copy where symlinks are unavailable). native means the tool reads .agents/skills itself, so nothing is created. — means the tool's project-level skill location is not implemented here.

Grok folder trust. Grok ignores a project's MCP servers and its skills until the folder is trusted, and says nothing about it: grok inspect simply lists none. Trust lives in ~/.grok/trusted_folders.toml, separate from config.toml. agents doctor reports an untrusted folder and agents doctor --fix records the trust.

Shared files. .mcp.json is read by both Claude Code and Copilot CLI, so agents writes it once for both. A server targeted at only one of them is still visible to the other; the sync says so rather than pretending targets isolate it.

Goose secrets. Goose reads credentials from the environment through env_keys, so values are never written into its config. The sync lists which variables have to exist in your shell.

Project Layout

your-project/
├── AGENTS.md                        ← Canonical instructions for all tools
├── CLAUDE.md                        ← Generated Claude wrapper (`@AGENTS.md`)
├── .agents/
│   ├── agents.json                  ← MCP servers, profiles, config (commit this)
│   ├── local.json                   ← Secrets & overrides (gitignored)
│   ├── mcp_config.json              ← Generated Antigravity workspace MCP
│   ├── skills/                      ← Reusable workflow definitions
│   │   └── my-skill/SKILL.md
│   └── generated/                   ← Auto-generated artifacts (gitignored)
│
│  ┌─── Generated by `agents sync` ───┐
├── .mcp.json                         │  Claude Code + Copilot CLI
├── .codex/config.toml                │  Codex
├── .grok/config.toml                 │  Grok Build
├── .gemini/settings.json             │  Gemini CLI
├── .cursor/mcp.json                  │  Cursor
├── .vscode/mcp.json                  │  Copilot VS Code
├── .amp/settings.json                │  Amp
├── .factory/mcp.json                 │  Factory Droid
├── .kilo/kilo.jsonc                  │  Kilo
├── .devin/mcp_config.json            │  Devin CLI
├── .zed/settings.json                │  Zed
├── opencode.json                     │  OpenCode
├── .junie/mcp/mcp.json               │  Junie
└── .claude/skills → .agents/skills   │  Skill bridges

Git strategy. In the default source-only mode, only .agents/agents.json, .agents/skills/ and AGENTS.md are committed: everything the sync generates is gitignored and rebuilt from them. In commit-generated mode the generated files stay out of .gitignore, so a clone gets working tool configs without installing this CLI. That is how you commit .mcp.json for the rest of the team.

Either way, files that hold a tool's own settings (.zed/settings.json, .amp/settings.json, .kilo/kilo.jsonc) and .github/mcp.json are never added to .gitignore: the sync merges its entries into them and leaves the rest of the file, including whether you track it, to you.

Command Overview

Setup & Sync

Diagnostics

MCP Servers

Profiles

Plugins

Skills & Integrations

Global Mode

Run any command with --global (or -g, or from $HOME) to manage MCP servers and skills machine-wide from ~/.agents/agents.json:

agents init -g
agents connect -g --llm opencode,codex,grok
agents sync -g

Tools that only have a user-level config (Claude Desktop, Devin Desktop, Goose) are written there in both modes.

Agent Plugins

Agent Plugins 1.0.0 is a vendor-neutral package format for skills and MCP servers, read by ChatGPT and Codex, Cursor, GitHub Copilot, Kiro and VS Code. A package is a directory with plugin.json, an mcp.json and a skills/ folder, which is close to what .agents/ already holds.

agents plugin export --name my-stack --plugin-version 1.0.0

dist/agent-plugin/
├── plugin.json      $schema, name, version, description
├── mcp.json         $schema, mcpServers
└── skills/          copied from .agents/skills

Export reads .agents/agents.json only, never local.json: what ships is the ${VAR} placeholder from the committed file, and the command prints which variables the package needs. ${PROJECT_ROOT} becomes the specification's ${PLUGIN_ROOT}.

Going the other way, agents plugin import ./some-plugin adds the package's servers under its plugin name (my-stack.docs) so nothing you already defined is replaced.

Context Budget

MCP servers cost context before an agent does any work: every tool definition is loaded up front.

agents mcp budget

MCP context budget:
  everything              13 tools  ~  1916 tokens
  filesystem              11 tools  ~  1240 tokens
  broken                 error: spawn broken-server ENOENT

Servers        3
Tools          24
Estimated      ~3156 tokens of context

The command speaks MCP 2026-07-28 directly over stdio and streamable HTTP, so the numbers come from the servers themselves. Servers built against the handshake revisions (2025-11-25 and earlier) are detected and handled the way the specification prescribes, so both eras work. Token counts are estimated from the size of the tool definitions, not measured by a model. Add --verbose for a per-tool breakdown, and --profile to measure what one profile would cost.

Servers whose config still contains an unresolved ${VAR} are skipped rather than started.

MCP Server Examples

# From a catalog URL
agents mcp add https://mcpservers.org/servers/context7-mcp

# stdio server
agents mcp add my-server --command npx --arg @my-org/mcp-server --arg /path/to/project

# HTTP server with a secret header
agents mcp add company-api --url "https://api.company.com/mcp" \
  --secret-header "Authorization=Bearer YOUR_API_TOKEN"

# Only for specific tools
agents mcp add ide-server --command ide-mcp --target cursor --target zed

A server definition supports timeout, connectTimeout, tools, disabledTools, oauth, headersHelper, bearerTokenEnvVar and envFile. Each is written only for the tools that document it, and the sync says when a field is dropped for a target that does not support it.

Values support ${VAR} and ${VAR:-default}, plus ${PROJECT_ROOT}.

Security

• Secret-looking values are moved to local.json and replaced with ${PLACEHOLDER} in agents.json when a server is added.
• agents doctor warns about literal secrets in committed config.
• Env keys and header names are validated before they reach a config file or a shell.
• agents plugin export never reads local.json.
• A config this CLI does not gitignore keeps the ${VAR} placeholder from agents.json, whether the value would have come from local.json or from your shell. Amp, Zed and Kilo share a file with the tool's own settings, and .github/mcp.json is a file teams review; the sync lists which variables to export so those tools can resolve them. commit-generated mode holds secrets back from every generated config for the same reason.
• A config this CLI rewrites keeps the permissions it had, so a file you restricted to 0600 stays that way.

Team Workflow

Lead sets up the project:

agents start
agents mcp add https://mcpservers.org/servers/context7-mcp
git add .agents/agents.json .agents/skills/ AGENTS.md && git commit -m "Add agents config"

In source-only mode that is everything the team needs: each clone runs agents sync and gets the tool configs locally. Switch syncMode to commit-generated if you would rather commit .mcp.json and the rest for people who do not install the CLI.

New member onboards:

git clone <repo> && cd <repo>
agents start        # keeps the committed config, writes local tool files

CI:

agents sync --check  # exits 2 when a tool config drifted from .agents/

FAQ

Documentation

• docs/agents-system.md — how sync, skills and generated files work
• docs/EXAMPLES.md — project setups
• CONTRIBUTING.md — development workflow
• CHANGELOG.md — release history

License

Apache 2.0. See LICENSE and NOTICE.