Pendraic for MCP users

Connect Pendraic to Claude Desktop, ChatGPT, Codex, or any MCP-capable client and run a full novel build, world expansion, or editorial pass on the AI subscription you already pay for. Premium tier feature.

What is the MCP server

MCP (Model Context Protocol) is the standard your AI host uses to talk to outside services. Pendraic ships an MCP server at mcp.pendraic.com that exposes every dashboard tool you already know: project create, worlds, story index, manuscript, anchors, voice, analysis, and the Editorial Workbench.

The difference: when you use the dashboard, Pendraic runs the AI for you and bills for tokens out of your wallet (Managed AI) or your BYOK key. When you use MCP, your host runs the AI on its own subscription. You get the structured layer; you bring the inference.

Why this matters

If you already pay for Claude Pro, ChatGPT Plus, Codex, or any other AI subscription, you’ve already paid for the inference. Doing a full novel build through the dashboard would charge you a second time. Going through MCP means you don’t double-pay. Your host subscription covers the LLM calls; Pendraic charges only for Premium access.

Connecting your host

From Account » AI Tools, mint an MCP token. Most MCP clients ask for two things: the server URL and a bearer token. Use:

  • Server: https://mcp.pendraic.com/api/mcp
  • Auth: Bearer (paste the token your dashboard generated)

Once connected, your host LLM sees Pendraic’s tools alongside its own. Ask it to “list my projects” or “draft chapter 3 of Riftforged” and it routes through Pendraic.

Atomic ops vs host-runnable plans

Pendraic’s tools split into two kinds.

Atomic opsare direct reads and writes. Create a chapter, add a story index entry, update a project’s logline, fetch the world’s bestiary. These run instantly and don’t involve any AI.

Host-runnable plans wrap the big workflows. Outline a novel from a logline, seed a world from a synopsis, draft a full chapter scene by scene, run the editorial full pass across eight layers. Your host fetches a plan envelope via workflow.plan, walks the steps, runs each prompt against its own LLM, and persists the output back through Pendraic’s atomic dispatcher writes.

Most modern MCP hosts handle this loop automatically. You ask for something complex; your host fetches the right plan; it runs the steps. You watch the outcome land in your dashboard.

What’s available

Every dashboard surface has an MCP equivalent. Project metadata, worlds and the worldbuilder, the story index, manuscript editing, beat maps and outlines, anchors and voice print, manuscript and world analysis, the Editorial Workbench, snapshots, exports. The long-form workflows you’d run from the dashboard’s bookshelf (outline, full novel pipeline, series pipeline, saga pipeline, editorial full pass, consistency audit) all have host-runnable plan kinds.

The only exception is cover image generation. Most MCP hosts don’t render images, so cover work stays in the dashboard.

Watching progress

When your host walks a long-running plan, Pendraic mirrors the progress to your dashboard’s workflow dock so you can watch a saga build from either surface. The dock won’t show LLM cost (your host owns that) but it shows step status, duration, and any token counts your host chooses to attach.

Tier

The MCP server is a Premium tier feature. Free and Essentials tokens are rejected with a helpful upgrade message. If you downgrade after issuing a token, the server re-checks your tier on every call so the gate fires immediately, no token revocation needed.

Related