Integrations

Works with your AI coder.

PostMQ is one MCP server every client sees the same way. Four AI coders are the ones we run ourselves, daily; anything else that speaks MCP connects with its own client config and the same once-disclosed credential.

Get started The MCP server & tools 87 MCP tools · the same catalogue for every client

Claude Code → .mcp.json

Codex → ~/.codex/config.toml

ZCode → ~/.zcode/cli/config.json

Antigravity → ~/.gemini/config/mcp_config.json

any MCP client → its own config, same URL

The four we run

Claude Code, Codex, ZCode, Antigravity — daily

Founder-attested, not a logo wall: the four AI coders we run ourselves, daily, against the same MCP surface — including the hand-offs where one builds and another reviews. All four also take the session-state kit — Antigravity's install being the one that wires nothing: its skills run the scripts themselves.

Claude Code

This client is where this repository's session protocol was written: one postmq entry in .mcp.json, one restart, then the start-session and end-session skills and the hooks we run. The full setup, the skills and a session end to end are on the Claude Code page.

config:.mcp.json

Codex

Add one [mcp_servers.postmq] entry — url plus bearer_token_env_var, so the credential stays in the environment and never in the file — or run one codex mcp add. The same config serves the CLI, the IDE extension and the desktop app; the Codex page has the setup and a reviewer's session end to end.

config:config.toml

ZCode

ZCode reads MCP servers from ~/.zcode/cli/config.json, and inside it servers nest under mcp.servers rather than the top-level mcpServers block Claude Code and Antigravity share: a type, a url, an Authorization header, a timeoutMs. The kit installs here too, as skills under .zcode/skills and hooks in .zcode/config.json. The ZCode page has the setup and the three things its hook surface does differently.

config:config.json

Antigravity

Antigravity reads MCP servers from ~/.gemini/config/mcp_config.json — the same mcpServers shape as Claude Code's .mcp.json entry: the server name, the url, an Authorization: Bearer header. The kit installs here too, as skills under .agents/skills and scripts under .agents/tools — and it is the one install that wires nothing: its skills run the scripts themselves, so usage there is recorded only when end-session runs. The Antigravity page has the setup and why.

config:mcp_config.json

four clients, one server

.mcp.json · Claude Code

{
  "mcpServers": {
    "postmq": {
      "type": "http",
      "url": "https://mcp.postmq.com/",
      "headers": { "Authorization": "Bearer pmq_…redacted…" }
    }
  }
}

config.toml · Codex

[mcp_servers.postmq]
url = "https://mcp.postmq.com/"
# names the variable whose value becomes the Authorization: Bearer header —
# the credential stays in the environment, never in this file
bearer_token_env_var = "POSTMQ_CREDENTIAL"

config.json · ZCode

{
  "mcp": {
    "servers": {
      "postmq": {
        "type": "http",
        "url": "https://mcp.postmq.com/",
        "headers": { "Authorization": "Bearer pmq_…redacted…" },
        "timeoutMs": 90000
      }
    }
  }
}

mcp_config.json · Antigravity

{
  "mcpServers": {
    "postmq": {
      "url": "https://mcp.postmq.com/",
      "headers": { "Authorization": "Bearer pmq_…redacted…" }
    }
  }
}

The four connect to the same records over the same tools — a build session here, a review directive there — and they interoperate because none of it is client-specific: the protocol is session state plus handoffs, and the marquee pairing — one agent builds, another reviews — is multi-agent code review.

Everything else

Any MCP client connects

PostMQ speaks standard MCP. If your editor or agent can add an MCP server, it can add this one — with its own config file and the same credential.

Cursor configures MCP servers in mcp.json (project .cursor/ or global ~/.cursor/) with stdio, SSE and streamable HTTP transports and custom headers, per Cursor's MCP docs. VS Code configures them in .vscode/mcp.json with stdio and http types, per the VS Code MCP docs. GitHub Copilot uses MCP servers in the IDEs it ships in, per GitHub's docs. For each of these, configure the server from the vendor's own instructions with the URL and the bearer credential from the mcp_config form; we have not published a PostMQ-tested recipe for these clients, and this page will say so until one exists.

The floor under all of it

One server, one catalogue, one credential

Every client above sees the same 87 tools, because there is one MCP server: Streamable HTTP at mcp.postmq.com, and stdio when co-located. The credential is scoped and once-disclosed — issue it in the mcp_config form and the response carries the paste-ready block next to the plaintext, shown once. The catalogue, tool by tool and argument by argument, is the MCP server page; the three-step connect is the docs.

Last verified 2026-09-03 against main at c36f170: the Codex, ZCode and Antigravity config shapes against their sources (the official Codex MCP docs; a live ZCode config; a live Antigravity config), the vendor MCP documentation links by retrieval, and the founder attestation recorded in the claims register the same day.

Point your coder at it.

Get started Read the docs