subagentic.ai RSS X YouTube GitHub
AI-generated illustration

News

Tailscale Aperture shares agent skills over MCP

Tailscale’s November 10, 2026 Aperture update stores skills for a user or an organization and distributes them to compatible agents through Skills over MCP, with admin control over who receives them.

tailscaleaperturemcpskills

Tailscale is updating Aperture so a person or an organization can store agent skills in one place and hand them to compatible agents over MCP. Remy Guercio and Ondrej Bilek published the change on November 10, 2026. The same day, Tailscale’s account on X called Aperture “one home for your team’s AI skills,” with control over who can access them and distribution “to compatible agents through Skills over MCP.”

Skills usually stay attached to one user and one agent, so coworkers copy files to stay aligned. Tailscale compares that to hunting a spreadsheet named Q4REPORT_FINAL_V2_NEW_20261110.XLSX in a long thread, and says that kind of sync will not hold as AI moves. Aperture is the proposed replacement: a repository for a user’s skills across agents, and for organization-level skills shared across teams.

Users and admins can import existing skills four ways:

  • Upload a SKILL.md, a .zip, or pasted Markdown.
  • Paste a public GitHub repository link so Aperture imports the bundle.
  • Supply a URL for a third party that supports the .well-known skill-discovery specification. Tailscale’s example is stripe.com; Aperture then lists skills to import.
  • Add Aperture as an MCP server and ask the connected agent to add existing skills.

Skills can also be created in Aperture. Select Add from the Skills menu, or use the pre-built Skill Creator skill. Skill Creator uses new MCP tools on the Aperture MCP gateway, so an agent on Aperture’s MCP endpoint — or Aperture Chat — can create, update, and delete skills.

Aperture’s MCP endpoints can also list and read LLM sessions no matter which agent created them. The suggested loop is to use a skill, have another agent review those sessions, and ask for improvements, with the history kept in Aperture.

Distribution is identity-scoped, not a generic connector. Aperture implements the Skills over MCP specification and “delivers identity-based access to LLMs, MCP servers, and skills.” An individual can send skills across their own agents or computers. Admins can decide which skills go to which people or teams based on identity.

Not every agent supports the Skills extension yet. Tailscale says that as agents add support, Aperture works with them automatically. The post does not name current clients, admin-policy steps, or price.

Tailscale asks organizations trying this to use Aperture and send feedback to aperture@tailscale.com.

Read the November 10 post, then try one import path you already have — a file, a public repo, a skill-discovery URL, or an agent on Aperture’s MCP server — and check that your agents actually accept the Skills extension before you rely on the share.

Sources