StoryOS
Start free

AI · MCP

Use your own AI. We never meter it.

StoryOS ships a Model Context Protocol server. Point the Claude or ChatGPT you already pay for at your workspace and it works your real data — and your AI bill stays exactly where it is.

$0

what StoryOS charges when your agents run on your own Claude or ChatGPT — on every plan, including Free

2 min

to connect: paste one address, sign in with your StoryOS account, approve — no tokens, no config files

47

MCP tools — the agent gets the same surface you do, from queries to building whole databases and automations

Connect by signing in — no config

StoryOS is an OAuth connector: paste one address, sign in, approve. That's the whole setup.

In Claude

1

Open Connectors

In Claude (web or desktop): Settings → Connectors → Add custom connector.

2

Paste the address

https://mcp.storyos.dev/mcp — that’s the whole configuration.

3

Sign in and approve

Claude sends you through a normal StoryOS sign-in. Approve, and your workspace appears in Claude’s tools. No tokens, no JSON.

In ChatGPT

1

Turn on connectors

In ChatGPT: Settings → Connectors → Advanced → enable Developer mode (available on paid plans).

2

Add the server

Create a new connector with the address https://mcp.storyos.dev/mcp.

3

Sign in and approve

ChatGPT runs the same StoryOS sign-in. Approve, then enable the connector in a conversation to use your workspace.

Or let your assistant set itself up:

What gets copied 
You are connecting to a StoryOS workspace over MCP.

1. Preferred (no tokens): add the remote MCP server
   https://mcp.storyos.dev/mcp as a connector. It supports OAuth - the user
   signs in with their StoryOS account and approves access. In Claude this is
   Settings -> Connectors -> Add custom connector; in ChatGPT it is
   Settings -> Connectors (developer mode) -> Add.
2. Advanced fallbacks, only if the client cannot do OAuth connectors:
   - Remote with a token: url https://mcp.storyos.dev/mcp with header
     "Authorization: Bearer <mn_pat_ token>" (created in the StoryOS app
     under Settings -> API tokens).
   - Self-hosted: command "node packages/mcp/dist/index.js" with env
     STORYOS_URL=<their instance URL> and STORYOS_TOKEN=<token>.
3. Once connected, call the get_started tool first - it explains the
   workspace/space/database model and the query filter format.
4. Use describe_database before reading or writing any database so you use
   real field api_names. Never guess field names or IDs.
Advanced: Claude Code, tokens & self-host

For clients that don't do OAuth connectors — or for scripts and self-hosted instances — create a personal access token at Settings → API tokens and configure the server directly. Tokens are scoped read / write / admin, and the agent is only offered the tools its token allows.

Hosted endpoint, token auth

{
  "mcpServers": {
    "storyos": {
      "type": "http",
      "url": "https://mcp.storyos.dev/mcp",
      "headers": {
        "Authorization": "Bearer mn_pat_…"
      }
    }
  }
}

Self-hosted, local stdio server

{
  "mcpServers": {
    "storyos": {
      "command": "node",
      "args": ["packages/mcp/dist/index.js"],
      "env": {
        "STORYOS_URL": "https://your-instance",
        "STORYOS_TOKEN": "mn_pat_…"
      }
    }
  }
}
The StoryOS settings page for creating personal access tokens

Full walkthroughs in the MCP docs.

The toolset, documented

The agent gets the same surface you do — discover, read, write, and build.

Discover

The agent orients itself — no docs needed.

get_startedlist_workspaceslist_spaceslist_databasesdescribe_databasesearchlist_icon_setget_record_descriptionlist_sources

Read

Structured queries, not a made-up query language.

query_recordsget_recordget_linkslist_attachments

Monitor

Visibility into skills, runs and approvals — read-only by design, so the human stays the one who approves or reruns.

list_skillslist_approvalsget_runs

Write

Validated server-side — mistakes get named, not saved.

create_recordupdate_recordupdate_record_descriptiondelete_recordlink_recordsunlink_recordsadd_commentattach_filedelete_attachmentrun_buttonrun_skill

Automations

Your agent can build and edit automation rules, not just fill in the data they act on — admin-scoped tokens only.

list_automationsget_automationcreate_automationupdate_automationdelete_automation

Build

Agents can construct whole workspaces, not just fill them — admin-scoped tokens only.

create_spacecreate_databaseupdate_databasedelete_databaseadd_fieldupdate_fielddelete_fieldchange_field_typereorder_fieldscreate_viewupdate_viewdelete_viewreorder_viewscreate_relationdelete_relation

Full reference in the MCP docs.

Why it doesn't hallucinate your data

The agent reads a described schema, filters with structured queries instead of a made-up query language, and every write is validated server-side — so mistakes get named, not silently saved. Your databases are the agent's long-term memory; there's no context stuffing to drift.

Start free How agents run →

Your data, your AI, two minutes.

Start free — 30 days of Pro, no card. Connecting your own Claude or ChatGPT costs you nothing, ever.

Start free →