Connect an agent

OpenStoryHub speaks MCP: any client that can hold an OAuth token and call a tool can read, edit and draft here on a writer’s behalf, with their consent, under their name. The server is a thin wrapper over the same API a browser calls, and every tool below is a route a third-party developer could reach directly, so nothing an agent can do here is a capability that exists only for agents.

Every write carries two identities: who physically made it, and who it was made for. There is no anonymous save, and nothing an agent writes is unattributed. It is a permanent, append-only checkpoint with both names on it, visible in the manuscript’s own history.

Connecting a client

1. 1. Sign in, or create an account.

   An agent always acts on behalf of a specific writer; the consent screen in step 3 needs one to grant access.

   Sign inStart a story

2. 2. Point your client at this URL.

   Most MCP clients discover everything else from it: the authorization server, the scopes it can ask for, over /.well-known/oauth-protected-resource.

   https://openstoryhub.co/mcp

   Copy

   Claude Code

   claude mcp add --transport http openstoryhub https://openstoryhub.co/mcp

   Copy

   Then run /mcp in a session, select it, and choose Authenticate. That opens the consent screen below in a browser.

3. 3. Approve only what you mean to grant.

   The consent screen lists every scope the client asked for as its own checkbox. Grant a subset, or all of it. drafts:create and notes:write start unticked no matter what was asked for: adding new pages and speaking in someone else’s circle are their own decisions, not bundled into “can edit my drafts.”

4. 4. Install a skill, if you want one.

   A scope is what the agent may do. A skill is instructions for how you want it done: text the agent reads before it works, installed separately from the cast, and taking effect on its next call. Neither substitutes for the other: a skill with no matching scope has nothing to act with, and a scope with no skill installed is an agent working from its own defaults.

What a connection is granted, checked on every call. Refused with a 403, not promised in a skill’s text. Two are narrower than they look: bible:write reaches the story bible and never the manuscript, whatever else is held, and drafts:create makes a new scene and cannot touch one that already exists without drafts:write alongside it.

• 01
  stories:read

  Read published stories, including your own.

• 02
  stories:write

  Publish and edit stories in your name.

• 03
  drafts:read

  Read your drafts, including revisions you have never published.

• 04
  drafts:write

  Append revisions to your drafts, and retitle, archive or restore a scene that already exists. Each revision is filed under this agent and under you, permanently.

• 05
  bible:write

  Write to your story bible. It reaches your story bible and never the manuscript, whatever else is held.

• 06
  drafts:create

  Create a new scene in your stories. A separate decision from editing the scenes you already have.
  Off by default: a separate consent decision.

• 07
  circles:read

  Read the writing circles you belong to and the notes left in them.

• 08
  circles:write

  Create a circle, invite or remove members, and bring a sponsored agent seat into one, on your behalf.

• 09
  notes:write

  Leave notes on other writers’ revisions in your circles, signed by this agent acting for you.
  Off by default: a separate consent decision.

• 10
  notes:scene

  Leave paragraph-anchored notes on your scenes: what is wrong, and where. Never writes to the manuscript, and never appears in provenance.
  Off by default: a separate consent decision.

Read-only. Every scope below unlocks writing; these need only drafts:read or stories:read.

• 01
  search_stories

  Search stories by form or free text: published work, plus the circles this writer belongs to.

• 02
  list_my_stories

  The stories this writer owns or co-authors, published or not.

• 03
  list_scenes

  The drafts in a story: ids, labels, order, word counts, scenes and bible entries alike.

• 04
  get_draft

  A single draft and its most recent revision.

• 05
  compile_story

  A whole story as one document: every active scene, in order, in one call.

• 06
  search_in_story

  Find a word or phrase inside one story, without re-reading every scene.

• 07
  get_provenance

  Who wrote a draft: a count per contributor. Never a ratio or a score.

• 08
  get_revision_history

  Every revision of a draft, with the actor that made it.

bible:write reaches a `note` draft and never a `scene`, enforced at the database rather than promised in prose.

• 01
  create_note

  Needs bible:write

  Create an empty story bible entry.

Revising a scene and adding a new one are two different grants. See the scopes below.

• 01
  create_scene

  Needs drafts:create

  Create an empty scene at the end of a story. A separate grant from revising one that exists.

• 02
  propose_revision

  Needs drafts:write

  Append a revision to a draft: the actual write, permanent and attributed.

• 03
  rename_scene

  Needs drafts:write or bible:write

  Retitle, archive or restore a scene or bible entry. Records no authorship.

Off by default, on purpose. Critique under this writer’s name in someone else’s circle is a separate decision from editing their own draft.

• 01
  post_note

  Needs notes:write

  Leave feedback on a revision in a writing circle. Asks the writer to confirm before it posts, on a client that supports asking.

No scope required. A skill is instructions, and reading which ones exist is not a write.

• 01
  list_skills

  The skills this writer has installed on this connection.

• 02
  load_skill

  Fetch an installed skill in full, and the revision it was served from.

• 03
  list_catalog_skills

  Browse what exists and could be installed: summaries only.