Integrations
Connect an AI assistant with the MCP server
Connect an external AI assistant — such as Claude — to your SceneItAll project so it can read your scenes, breakdown, schedule, and reports, and help you work. This article covers how to connect, what the assistant can do once connected, and what MCP actually is.
Most tools are read-only, but the server also supports a set of mutation tools (scene updates, element tagging, availability changes, schedule edits). What the assistant may do is governed by a connector permission mode you choose. By default it’s Ask before each operation — the assistant can act, but checks with you first. See permission modes below.
Character creation through MCP uses the same project-unique Character-number allocator as the app, and Cast inherits the one assigned Character number. The fill_availability_gaps mutation fills only uncovered time as Available or Unavailable and preserves all existing slots.
MCP is the bridge between your AI assistant and SceneItAll
The Model Context Protocol (MCP) is an open standard that lets an external AI assistant call a defined set of tools on another system. SceneItAll’s MCP server is the bridge: when you ask the assistant to do something, it calls a SceneItAll tool, and the server turns that call into an API request using your authenticated session.
Connecting is per-user and authenticated. The server runs every request as you — so the assistant sees only the projects you’re a member of and can take only the actions your role permits.
Before you start
- The assistant connects to the projects you can access, at your role. It can do only what your role allows in the app.
- You sign in with your SceneItAll account during setup. You never paste a password into the assistant — you authenticate through the normal SceneItAll sign-in.
- This build of SceneItAll must have the MCP server configured. If it is not, the AI Connector settings panel shows a notice instead of a server URL.
Get your server URL from AI Connector settings
To find your server URL, open the AI Connector settings in the SceneItAll app:
- In the top bar, select your avatar.
- Choose MCP Connection from the menu. This opens Settings → Account → AI Connector — the single home for the connector (you can also reach it from Profile Settings, then Account → AI Connector).
- Under Connect an AI assistant, copy the value below MCP Server URL with the Copy button.
Tip: the panel also shows a ready-to-paste Claude Web setup snippet. The Copy button beside it copies the same server URL to your clipboard.
[SCREENSHOT: The AI Connector settings panel showing the MCP Server URL field with a Copy button, the Claude Web setup snippet, and the connector access level dropdown]
If the panel shows the message that MCP is not configured for this build, the server URL is unavailable and you cannot connect from this app until it is set up.
Add SceneItAll to Claude as a custom connector
To connect Claude on the web, add SceneItAll as a custom connector:
- In Claude, go to Settings, then Connectors, then Add.
- For the name, enter
SceneItAll. - For the remote MCP server URL, paste the MCP Server URL you copied.
- Select Add.
When you connect, you are prompted to sign in with your SceneItAll account, as the panel notes: “You’ll sign in with your SceneItAll account when connecting.” After you sign in, the assistant can act on the projects you can access.
Set your context, then ask for what you need
Once connected, the assistant sets your working context before it does anything else, then helps you from there:
- It lists your projects. One project is selected automatically; with several, it asks you to choose.
- It does the same for the episode. A script revision is selected only when you ask for revision-native work such as comparison, translation, or continuity.
- With the project and episode set, ask for production work such as scene breakdowns, cast and crew availability, schedule details, call sheets, and reports. Those tools use stable episode scenes and the latest production-ready revision automatically.
- Optionally, apply a persona (see below) to focus the assistant on one production role.
- What the assistant may do depends on your connector’s permission mode. By default (Ask before each operation), it proposes each step and waits for your “yes” before running it; in Full access it acts without asking; in Read-only every change is blocked; in Blocked it can’t do anything (see permission modes).
- If you do not want the assistant to receive screenplay dialogue or action, enable Script content blocked in the same AI Connector panel. Production metadata remains available.
Talk about days and scenes the way you normally would. Say “Day 3” or “scene 24A” and the assistant resolves that to the actual record before acting — it never guesses or acts on a number that could mean two different things. Day numbers can repeat across production units (a Main Unit “Day 3” and a 2nd Unit “Day 3” are different days). Production-scene numbers resolve within the active episode and use the latest production-ready details; revision-specific script tools still use the explicitly selected revision.
What the assistant can do
The MCP server gives the assistant a defined set of tools. Most are read-only; the rest are mutation tools that make changes, gated by your connector’s permission mode (see permission modes). The tools span these areas:
| Area | What the assistant can do |
|---|---|
| Project context | List your projects, set the active project and episode, optionally select a revision for script-native tools, and read a project overview or health dashboard. |
| Scripts and scenes | List episodes and scripts; list the active episode’s latest production-ready ordinary scenes, copies, and split parts; read scene detail and comments; add notes to the scene Notetaker; group scenes by set or location; update scene fields; and create a light or full independent production-scene copy with an optional reason. |
| Elements and breakdown | List and inspect elements, read a scene’s breakdown and the cast list, create and recategorize elements (including project-specific custom categories), tag or untag elements on scenes, and permanently delete duplicate or obsolete elements. |
| Crew, cast, and availability | List crew and cast, search people by name, read the monthly availability grid, and set availability for a person, a date range, or several people at once. |
| Locations | List filming locations, read location detail, find the scenes at a location, and create or edit a location. |
| Conflicts | List scheduling conflicts and check what conflicts a proposed schedule change would cause. |
| Schedule | Read the schedule, one-liner, strips, scenarios, and schedule settings; compare two scenarios; and create or edit scenarios, shoot days, strips, banners, and schedule settings. |
| Call sheets | Read a day’s call sheet, set its weather note, refresh it from upstream, and export it to PDF. |
| Reports | Read the Day Out of Days, the full-fat schedule, a single element’s Day Out of Days, and continuity warnings. |
| Data completeness | Scan the whole project for fields you haven’t filled in yet — like character prep times that would otherwise fall back to call-sheet defaults — grouped by area, so you can fill the ones that matter. |
| Search | Run a full-text search across the active project and read that project’s recent notifications plus platform-wide messages. |
The exact tool set is defined in the MCP server (SceneItAll_App/mcp_server/server.py).
Check what you haven’t filled in
The assistant can audit your whole project for blank fields that are worth filling — so nothing important is missing from your reports. A common example is character prep times: if you never set a character’s wardrobe or makeup prep minutes, the call sheet quietly falls back to default times, and you might not notice.
Ask the assistant something like “What haven’t I filled in?” and it scans every part of the project — cast, crew, elements, scenes, locations, schedule, and reports — then lists the empty fields, grouped by area, with a note on why each one matters. You can ask it to fill specific values (for example, “set Jane’s makeup prep to 75 minutes”); if your permission mode allows, it makes the change and reminds you to refresh any affected call sheet so the new value shows up.
The audit only ever reports which fields are empty — it never shows the personal or financial data that MCP always keeps private (see What it never returns).
What it never returns
Some data is never exposed to the assistant, regardless of your role:
- Personal contact details, addresses, and agency information.
- Cast measurements and financial data.
You can also enable Script content blocked. While enabled, SceneItAll removes raw screenplay text, structured screenplay blocks, and translated screenplay content before an MCP tool receives the response. Scene numbers, sluglines, synopsis, breakdown elements, page counts, schedules, and reports remain available.
You decide what the assistant is allowed to do
Set the mode in Settings → Account → AI Connector. It’s enforced by the server itself — not just a polite request to the AI — applies to every project, and a change takes effect within a few seconds with no need to reconnect.
| Mode | What the assistant can do |
|---|---|
| Full access | Read everything and make changes (the mutation tools) with no per-step confirmation. For when you trust the assistant to run multi-step build/scheduling workflows end to end. |
| Ask before each operation (default) | The assistant can act, but checks with you before every operation — it proposes exactly what it’s about to do and waits for your “yes” before running it (see below). |
| Read-only | Browse, search, and read freely. Every change is blocked — safe exploration. |
| Blocked | The connector is off: the assistant can’t read or change anything. |
If the assistant tries something the current mode doesn’t allow, it gets a clear message and will tell you to change the mode in Settings. A brand-new connector starts in Ask before each operation, so the assistant always checks with you until you give it more (Full access) or less (Read-only / Blocked).
Keep screenplay text out of MCP
Enable Script content blocked below the access-level dropdown to prevent the connector from returning screenplay dialogue, action, structured script blocks, or translations. This setting is independent of the permission mode: for example, Full access can still update production data while screenplay content remains blocked. It applies across every project and takes effect within about 15 seconds without reconnecting.
The setting is off by default so existing connections keep their current behavior. When it is on, scene-detail results explicitly say that script content is blocked while still returning the scene’s allowed production metadata.
How “Ask” works. SceneItAll can’t pop up a window inside your AI app, so the approval happens in your chat: when the assistant wants to do something, it shows you exactly what (e.g. “set scene 5 to NIGHT?”) and waits. Only after you say yes does it actually run that step. Nothing happens until you approve — and because it covers every operation, expect to approve routine lookups too, not just changes.
Tip: to stop an assistant mid-task, switch the connector to Read-only or Blocked — it takes effect within ~15 seconds.
It only ever sees what you can see
- Your permissions apply. The MCP server uses your token, so it can only access projects you’re a member of, at your role level.
- You choose what it can do. A per-user permission mode (Full access / Ask before each operation / Read-only / Blocked) is hard-enforced by the server. Ask (the default) makes the assistant check with you before every operation; Read-only lets it browse but blocks every change; Blocked turns it off entirely. Script content blocked independently withholds screenplay text. Change both controls in Settings → Account → AI Connector.
- No personal data exposed. Contact details, addresses, agency info, measurements, and financial data are never returned by MCP tools.
- Tokens are never stored in the clear. Raw bearer tokens are never written to disk or logged; only a one-way SHA-256 hash is kept server-side to validate sessions, and the MCP process holds tokens in memory only.
- Rate limited. The same API rate limits apply to MCP requests.
Changes respect the data flow
SceneItAll data flows downstream only: Script → Breakdown → Crew, Departments, and Cast → Schedule → Reports. The assistant’s tools follow the same rules the app does. Revision selection affects script-native tools only. Production tools edit stable episode scenes and shared downstream data while reading screenplay fields from the latest production-ready revision. Reports do not feed anything upstream: for example, refreshing a call sheet re-pulls the current upstream schedule, it does not write back to the schedule. Whether the assistant may make those edits at all depends on your connector’s permission mode.
Apply a persona to focus on one production role
A persona is a starting-point prompt — not a SceneItAll role, and it doesn’t change your permissions. It shapes how the assistant prioritizes work and what it flags. The server offers five:
| Persona | Focuses the assistant on |
|---|---|
first_ad | Schedule integrity, conflicts, turnaround, crew logistics, and shoot-day planning. |
script_supervisor | Continuity, scene ordering, script-day logic, and time-of-day analysis. |
line_producer | Cost-aware scheduling, hold-day analysis, overtime prevention, and crew logistics. |
location_manager | Permits, location grouping, company moves, travel logistics, and weather contingency. |
casting_director | Cast availability windows, hold days, actor scheduling conflicts, and deal-aware scheduling. |
Apply a persona in your assistant the way it offers prompts, then ask your question as usual.