Developers

Build agents that compose decks against your firm's brand.

Codexroom exposes its workspace through a Model Context Protocol server and a REST API. Your engineering team wires it into Claude, Cursor, the OpenAI Agents SDK, or your internal stack. The reference below covers every endpoint and every tool.

What it looks like

Five tool calls from prompt to share.

The agent reads your tokens, searches your data room, composes inside a template, and ships a watermarked link. The validation engine runs before the deck ever leaves the workspace.

MCP session

ts
// An MCP client, connected over a scoped OAuth grant:

await resolve_tokens(["color.brand", "font.display"])
await search_assets("metric", { label: "active enterprise customers" })

await create_deck({
  name: "Acme — Q3 review",
  deck_template_id: "qbr-deck",
  slide_instances,
})            // runs validate_deck, returns the deck

await create_share({
  deck_id,
  recipient_name: "Acme Corp",
  gates: ["password", "email"],
})            // returns the watermarked share URL

What you can build

Three patterns most teams ship in the first quarter.

An agent that drafts decks for sellers.

Wire Claude Desktop into the workspace. The seller prompts, the agent composes, the link arrives in Slack.

A pipeline that ships quarterly updates.

Schedule a job that pulls the latest metrics, composes the quarterly LP letter from the template, and creates per-LP shares.

A copilot inside your existing app.

Embed the API into the internal tool your team uses every day. The deck assembly stays in their workflow.

Built for production

Scoped auth. Generous limits.

OAuth for agents. Service tokens for servers.

OAuth 2.0 with PKCE for MCP clients. Service tokens for server-to-server calls. Both scope-gated to the workspace, the user, and the operations you authorise.

Generous limits by default.

Published per endpoint, with overage rates on the pricing page. Enterprise contracts include custom limits.