MCP: drive bootload from your AI assistant

bootload ships an MCP server, so Claude Code, Claude Desktop, Cursor, Codex and every other MCP-capable client get the platform as native, typed tools — deploy, logs, metrics, wallet, domains, backups, tickets — instead of shelling out to the CLI.

It is built into the CLI you already have. One line of setup:

claude mcp add bootload -- bootload mcp

That is everything: bootload mcp serves MCP on stdio and reuses your stored bootload login (or BOOTLOAD_TOKEN / --token for a scoped API key — recommended for agents: bootload token create --scope projects:read --scope services:write --scope deployments:write).

Other clients

Claude Desktop / Cursor (JSON config):

{
  "mcpServers": {
    "bootload": { "command": "bootload", "args": ["mcp"] }
  }
}

Codex (~/.codex/config.toml):

[mcp_servers.bootload]
command = "bootload"
args = ["mcp"]

Don't have the CLI yet? curl -fsSL https://bootload.io/v1/cli/install.sh | sh

What the agent can do

Read-only tools (annotated as such, safe to call freely): projects, services, logs, metrics, wallet balance, usage, capabilities, domain search, tickets, app catalog. Mutating tools mirror the CLI's guarantees: errors carry the same stable {error, code} envelope and the production guard applies unchanged. Money-spending tools are double-gated: they refuse outright unless you started the server with BOOTLOAD_MCP_ALLOW_SPEND=1 (your standing policy), and even then each call requires confirm: true (the per-call check). Without the opt-in your agent can read, wait, tail logs and roll back — but never spend a cent. Enable it per client, e.g.:

claude mcp add bootload --env BOOTLOAD_MCP_ALLOW_SPEND=1 -- bootload mcp

Start any session with capabilities_list — it is the machine-readable map of what's stable, what costs money, and what's reversible.

No local install? Use the hosted endpoint

Web-based clients (and anywhere you can't run a local process) can speak MCP straight to the API — streamable HTTP at:

https://my.bootload.io/v1/mcp

Authenticate every request with your API key: Authorization: Bearer <key> (mint one with bootload token create or on the portal's account page). The same double gate applies: money-spending tools additionally need the X-Bootload-Allow-Spend: 1 header on your session. Scopes, the production guard and the audit trail apply exactly as everywhere else.

The full tool list grows toward CLI parity; agents that prefer plain text can keep using llms.txt and the deploy skill — same platform, same guarantees.