Skip to content
MCP-first, model-neutral

Keep your favorite AI. Give it a canon.

Alcove connects the writing assistant you already use to durable story state, structured checks, and a human Canon Lock—without asking you to move into another app.

Remote MCP · OAuth 2.1 · organization scoped
Your writing assistant

“Draft the next chapter from the approved direction. Keep the archive rule intact, and do not resolve Yren’s suspicion yet.”

  1. 1
    get_canon_context

    Your agent receives the current canon, open threads, and next direction.

  2. 2
    submit_proposal_version

    The chapter arrives as a proposal. It still cannot change canon.

  3. 3
    get_proposal

    Checks and review notes settle around the exact version submitted.

  4. 4
    lock_proposal

    Only your explicit Canon Lock makes the approved changes true.

Remote MCP server

https://mcp.alcove.fi/mcp
OAuth sign-inOrganization scopedConsent before Canon Lock
One spine, many writing surfaces

The conversation can move. The story state does not.

A chat thread is temporary. Your Save File is not. Return in a new session, switch assistants, or paste from a local model—the same locked canon remains authoritative.

Full loop

MCP-capable assistants

Connect the remote server, sign in with Alcove, and let your assistant read canon, submit chapters, inspect checks, and ask before Canon Lock.

Plan dependent

ChatGPT and other assistants

Use write-capable MCP where your plan supports it. Direct connector availability changes by host; the paste path always remains available.

Universal

Any model, local or hosted

Copy the canon context, write wherever you want, and paste the finished chapter back into the same checked review flow.

The agent loop

Context in. Chapter back. Checks before truth.

The MCP tools make the controlled workflow explicit. An assistant can help at every step, but it cannot silently bypass the review gate.

  1. 01
    get_canon_context

    Receive a bounded writing packet

    Locked canon, current cast and relationships, open threads, route pressure, and the recommended direction arrive together.

  2. 02
    submit_proposal_version

    Attach the chapter as a proposal

    The prose is stored as a reviewable version. Alcove extracts proposed canon changes when the assistant does not provide them.

  3. 03
    get_proposal

    Wait for checks and review notes

    The assistant can poll the exact proposal version until its checks settle, then explain the result in the conversation.

  4. 04
    lock_proposal

    Ask the writer before Canon Lock

    Canon Lock is the product-truth mutation. It refuses while checks are pending or failed and remains an explicit consent moment.

Designed for agents, controlled by writers

Useful access without giving away the keys.

Account, canon, and billing stay first-party. Connections inherit the signed-in organization and every consequential write goes through the same domain rules.

OAuth, not copied secrets

Connect through a browser sign-in. No Alcove API key needs to live in a chat configuration.

Organization-scoped reads

Every tool resolves access from the signed-in identity instead of trusting a client-supplied organization.

One Canon Lock service

Console, MCP, and REST calls converge on the same guarded transaction—never a direct canon write.

Producer-neutral by design

The canon and check model does not depend on which company or model produced the chapter.

Alcove REST API
Preview
POST /v1/saves/{saveId}/proposals
GET  /v1/saves/{saveId}/proposals/{number}
POST /v1/saves/{saveId}/proposals/{number}/lock

Idempotency-Key: <required on writes>
Authorization: Bearer <organization identity>
The same spine over REST

MCP is the first client. The contract is bigger than one host.

The versioned REST surface and generated OpenAPI schema expose the same Save File, proposal, check, and Canon Lock model for authenticated integrations.

The external developer platform—public API keys, SDKs, and a marketplace—is not being claimed here. Today’s API is the authenticated product spine and an integration preview.

View the generated API index

Use an MCP connection

Best when your assistant supports remote, write-capable MCP tools.

Connection guide

Paste from anywhere

The universal path for ChatGPT, local models, custom prompts, and human drafts.

Read the quickstart

Build against the API

Inspect the current REST/OpenAPI contract before planning an authenticated integration.

API index
Your assistant, your canon

Let the conversation stay fluid. Keep the story grounded.

Connect over MCP or bring the next chapter back by hand. Both arrive at the same checks and Canon Lock.