Coding agents and editors

Post to social media from Zed

Setup checked September 22, 2026Zed Industries

Zed is a code editor with an AI agent built in, and once you connect iHatePosting that agent can write and schedule your social posts. Zed calls MCP servers context servers, and a remote one is two fields in your settings file: a url and an Authorization header. Ask the agent to write a post about the change you have shipped, check it against each network's rules and schedule it.

Set it up

  1. 1. Create an iHatePosting API key

    Sign in, open Settings → Developers and create a key. It starts with pk_live_ and is shown in full only once, so copy it before you leave the page.

  2. 2. Add iHatePosting to Zed

    Add to ~/.config/zed/settings.json, inside its existing top-level object. On Windows that file is %APPDATA%\Zed\settings.json. The zed: open settings file command opens it. You can also use Settings → AI → MCP Servers → Add Server → Add Remote Server: name it ihateposting, paste the URL, and under Headers add Authorization with the value Bearer followed by your key. Replace YOUR_API_KEY with your key.

    {
      "context_servers": {
        "ihateposting": {
          "url": "https://ihateposting.com/mcp",
          "headers": {
            "Authorization": "Bearer YOUR_API_KEY"
          }
        }
      }
    }
  3. 3. Ask it to post

    Tell Zed what you want to say and where. It saves a draft in iHatePosting unless you ask for the post to go out now or at a set time.

What is different about Zed

  • Zed calls MCP servers context servers, so the block goes under context_servers, not mcpServers. Open the file with zed: open settings file from the command palette and paste the entry inside the existing top-level object — two top-level objects in one file is not valid JSON. You can also add it under Settings → AI → MCP Servers with Add Server, then Add Remote Server; Zed's docs say that route writes the entry into that same settings file.
  • Keep the Authorization header. Zed's docs say that when a remote MCP server has no configured Authorization header, Zed sends you through the standard MCP OAuth flow. Our key address takes the key in the header instead, and publishes no OAuth metadata for that path, so there is nothing for that flow to find. Without the header our server has no key to forward, and a tool call answers that no key was sent.
  • A green dot is not a working key. Zed's check is Settings → AI → MCP Servers, where a running server shows a green indicator whose tooltip says Server is active. Our endpoint answers the tool list without any credential, so it goes green even when the key is wrong, and only the tool calls fail, with the text iHatePosting API 401. Ask Zed which iHatePosting account it is using — the whoami tool returns the login your key belongs to. If it is not yours, create a new key in Settings → Developers.
  • Zed asks before each tool call. From Zed v0.224.0 approval is controlled by agent.tool_permissions.default, which is confirm and prompts before any tool action, including MCP tool calls. To let one of ours run without the prompt, add "mcp:ihateposting:create_post" under agent.tool_permissions.tools in the same settings file, with "default": "allow" — the ihateposting part is the name you gave the entry under context_servers, so change it there and the key changes with it.
  • Pattern rules will mostly miss. Zed's docs say pattern-based rules match against an empty string for MCP tools, so the default key on the per-tool entry is the primary mechanism for them. The permission prompt agrees — Zed's docs say MCP tools only support the tool-level option, so you are not offered Always for <pattern>.
  • Put the entry in your user settings, not in the project. Zed opens every folder in Restricted Mode, which stops a project's .zed/settings.json from installing and spawning MCP servers until you trust that folder — an untrusted folder shows an exclamation mark in the title bar. Servers installed globally are started as usual, independent of worktree trust, so a user-settings entry works in any folder and keeps your key out of your repository.

Limits. Zed's plan table lists no MCP row: every plan, Free included, has AI via your own API keys, and only Zed-hosted models are reserved for Pro, Student and Business. Zed Business admin controls cover collaboration, hosted models, edit predictions and feedback; the Admin Controls page lists no control over MCP servers, and says the controls do not cover third-party extensions, External Agents or Terminal Threads.

Zed Industries's own documentation: Model Context Protocol in Zed

Nothing goes out unless you say so

Every post Zed writes is saved as a draft in iHatePosting unless you ask for it to be published now or at a set time. Publishing, rescheduling, retrying and deleting are marked as actions that change things, so clients that honour those markings ask you before they run them. Deleting a post in iHatePosting never removes one that is already live on a network.

Frequently asked questions

Where does the iHatePosting entry go?

Under context_servers in your Zed settings file. Run zed: open settings file from the command palette and add the entry inside the existing top-level object. The route through the interface is Settings → AI → MCP Servers, then Add Server and Add Remote Server, which Zed's docs say writes the entry into that same settings file.

The dot next to the server is green, so why does nothing post?

Green means the server answered, not that your key is right. Our endpoint answers the tool list without any credential, so a wrong key shows the same green dot and only the tool calls fail, with the text iHatePosting API 401. Ask Zed which iHatePosting account it is using: the whoami tool returns the login your key belongs to. Create a fresh key in Settings → Developers if it is not yours.

Why does Zed ask me before every post?

Tool approval is set to confirm by default from Zed v0.224.0, for any tool action including MCP tool calls. To stop the prompt for one tool, add "mcp:ihateposting:create_post" under agent.tool_permissions.tools in the same settings file, with "default": "allow". Pattern rules will mostly miss, because Zed's docs say pattern-based rules match against an empty string for MCP tools.

Do I have to trust the folder first?

Not for an entry in your own settings file. Restricted Mode holds back a project's .zed/settings.json until you trust that folder, while globally configured MCP servers start as usual. Trust only comes into it if you put the entry in .zed/settings.json inside a project, which also puts your key in your repository.

Ready to post everywhere from Zed?

Start your 90-day free trial — no credit card, about a minute to your first post.

Sign Up

Other AI agents

See every AI agent