cd /news/ai-tools/microsoft-agent-package-manager · home topics ai-tools article
[ARTICLE · art-37527] src=github.com ↗ pub= topic=ai-tools verified=true sentiment=↑ positive

Microsoft Agent Package Manager

Microsoft released an open-source dependency manager for AI agents called Agent Package Manager (APM), which allows developers to declare agentic dependencies in a manifest file and reproduce them across different coding agents. The tool supports transitive dependency resolution, plugin authoring, and integration with multiple AI coding assistants including GitHub Copilot, Claude Code, and Cursor.

read5 min views1 publishedJun 24, 2026
Microsoft Agent Package Manager
Image: source

An open-source, community-driven dependency manager for AI agents.

Think package.json

, requirements.txt

, or Cargo.toml

— but for AI agent configuration.

GitHub Copilot · Claude Code · Cursor · OpenCode · Codex · Gemini · Windsurf · Kiro

** Documentation** ·

·

Quick Start·

CLI Reference

Roadmap

Portable by manifest. Secure by default. Governed by policy.One file describes every agent's context; one command reproduces it everywhere; one policy controls what an org will allow.

AI coding agents need context to be useful — standards, prompts, skills, plugins — but today every developer sets this up manually. Nothing is portable nor reproducible. There's no manifest for it.

APM fixes this. Declare your project's agentic dependencies once in apm.yml

, and every developer who clones your repo gets a fully configured agent setup in seconds — with transitive dependency resolution, just like npm or pip. It's also the first tool that lets you author plugins with a real dependency manager and export standard plugin.json

packages.

name: your-project
version: 1.0.0
dependencies:
  apm:
    - anthropics/skills/skills/frontend-design
    - github/awesome-copilot/plugins/context-engineering
    - github/awesome-copilot/agents/api-architect.agent.md
    - microsoft/apm-sample-package#v1.0.0
  mcp:
    - name: io.github.github/github-mcp-server
      transport: http   # MCP transport name, not URL scheme -- connects over HTTPS
git clone <org/repo> && cd <repo>
apm install    # every agent is configured

Coming from npx skills add? Drop-in:

apm install vercel-labs/agent-skills                            # whole bundle, like npx skills add
apm install vercel-labs/agent-skills --skill deploy-to-vercel   # one skill, persisted to apm.yml

Same install gesture. You also get a manifest, lockfile, and reproducibility.

Zero-config Copilot:

apm compile -t copilot   # writes .github/copilot-instructions.md

One command, no configuration -- VS Code and GitHub Copilot read the file automatically. APM dogfoods this target on its own repository.

One apm.yml

describes every primitive your agents need — instructions, skills, prompts, agents, hooks, plugins, MCP servers — and apm install

reproduces the exact same setup across every client on every machine. apm.lock.yaml

pins the resolved tree the way package-lock.json

does for npm.

— declared once, deployed across Copilot, Claude, Cursor, OpenCode, Codex, Gemini, Windsurf, KiroOne manifest for everything— GitHub, GitLab, Bitbucket, Azure DevOps, GitHub Enterprise, Gitea, Gogs, any git hostInstall from anywhere— packages can depend on packages; APM resolves the full treeTransitive dependencies— build Copilot, Claude, and Cursor plugins with dependency management, then export standardAuthor pluginsplugin.json

— install plugins from curated registries in one command, deployed across all targets and lockedMarketplacesPack & distributeapm pack

bundles your configuration as a zipped package or a standalone plugin— GitHub Action for automated workflowsCI/CD ready

Agent context is executable in effect — a prompt is a program for an LLM. APM treats it that way. Every install scans for hidden Unicode that can hijack agent behavior; the lockfile pins integrity hashes; transitive MCP servers are gated by trust prompts.

Content securityapm install

blocks compromised packages before agents read them;apm audit

runs the same checks on demand—Lockfile integrityapm.lock

records resolved sources and content hashes for full provenance—SBOM exportapm lock export --format cyclonedx|spdx

emits a standard inventory of what reached disk, straight from the lockfile — provenance for procurement, not a compliance attestation—Drift detectionapm audit

rebuilds your agent context in scratch and diffs it against your working tree to catch hand-edits before they ship— transitive MCP servers require explicit consentMCP trust boundaries

apm-policy.yml

lets a security team say "these are the only sources, scopes, and primitives this org will allow" and have every apm install

enforce it — with tighten-only inheritance from enterprise to org to repo, a published bypass contract, and audit-mode CI gates.

apm-policy.yml governs what gets installed; your agent harness governs what runs. The two planes do not overlap.

— the canonical enterprise reference: enforcement points, bypass contract, air-gapped story, failure semantics, rollout playbookGovernance Guide— every check, every field, every defaultPolicy reference— staged rollout from warn to block across hundreds of reposAdoption playbook— wireGitHub rulesets integrationapm audit --ci

into branch protection

curl -sSL https://aka.ms/apm-unix | sh
irm https://aka.ms/apm-windows | iex

Native release binaries are published for macOS, Linux, and Windows x86_64. apm update

reuses the matching platform installer.

Then start adding packages:

apm install microsoft/apm-sample-package#v1.0.0

Or install from a marketplace:

apm marketplace add github/awesome-copilot
apm install azure-cloud-development@awesome-copilot

Or add an MCP server (wired into Copilot, Claude, Cursor, Codex, OpenCode, Gemini, Windsurf, and Kiro):

apm install --mcp io.github.github/github-mcp-server --transport http   # connects over HTTPS

See the ** Getting Started guide** for the full walkthrough.

agentrc analyzes your codebase and generates tailored agent instructions — architecture, conventions, build commands — from real code, not templates.

Use agentrc to author high-quality instructions, then package them with APM to share across your org. The .instructions.md

format is shared by both tools — no conversion needed when moving instructions into APM packages.

Created by @danielmeppiel. Maintained by @danielmeppiel and @sergio-sisternes-epam.

Roadmap & DiscussionsContributingAI Native Development guide— a practical learning path for AI-native development

Built on open standards: AGENTS.md · Agent Skills · MCP

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.

── more in #ai-tools 4 stories · sorted by recency
── more on @microsoft 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/microsoft-agent-pack…] indexed:0 read:5min 2026-06-24 ·