for_agents_/mcp

The studio,readable fromClaude Code.

One endpoint over MCP. Claude Code reads the screenplay, the breakdown, the shot list, the Director's briefs, the art and every render. It changes nothing.

Connect Claude Code See the 16 tools

Streamable HTTP·stateless·MCP 2025-11-25·read-only

endpointhttps://stanislavsky.ai/api/mcp

POST · JSON

tools, every one of them a reader

writes, renders or spends

20,000

tokens one answer may carry

How to connect

One endpoint. One command.

The steps change with the client you pick at the top of the page. Everything runs on one key, and it opens only the productions your account can open.

Change client

Claude Code

One command adds the studio as a server, in the terminal you already work in.

HTTP · bearer

1
Make a key
In your profile, under API keys, press New API key. It is shown once — copy it then.

Add the server

Run this once. The key goes in the header, not in the URL.

terminal

claude mcp add --transport http stanislavsky https://stanislavsky.ai/api/mcp --header "Authorization: Bearer sk-stan-live-…"

3
Ask about a production
Claude lists the projects the key can open first, then reads the one you name.
a first question
```
Where does LURIA stand — and what is still unrendered, scene by scene?
```

Open your API keys

Cursor

A remote server in Cursor's own config file; the agent panel picks it up on reload.

HTTP · bearer

1
Make a key
In your profile, under API keys, press New API key. It is shown once — copy it then.

Add the server to .cursor/mcp.json

In the project you want it in — or ~/.cursor/mcp.json for every project.

.cursor/mcp.json

{
  "mcpServers": {
    "stanislavsky": {
      "url": "https://stanislavsky.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer sk-stan-live-…"
      }
    }
  }
}

3
Reload the window
The server appears under Settings → MCP with its tools listed. Ask in the agent panel.

Open your API keys

VS Code

GitHub Copilot's MCP config, with the key asked for once and kept by the editor.

HTTP · bearer

Add it to .vscode/mcp.json

The input below asks for the key the first time the server starts, and VS Code keeps it.

.vscode/mcp.json

{
  "inputs": [
    {
      "id": "stanislavsky-key",
      "type": "promptString",
      "description": "Stanislavsky API key",
      "password": true
    }
  ],
  "servers": {
    "stanislavsky": {
      "type": "http",
      "url": "https://stanislavsky.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:stanislavsky-key}"
      }
    }
  }
}

2
Start it
Run MCP: List Servers, choose stanislavsky, then Start. Its tools are available in Copilot Chat's agent mode.

Open your API keys

Other

The whole server is one POST: JSON-RPC in, one JSON body out. Any client that can send a header can use it.

HTTP · bearer

1
Make a key
In your profile, under API keys, press New API key. It is shown once — copy it then.
2
Send a call
Export the key once in your shell; every call carries it and stands alone, so any instance can answer it.
terminal
```
export STANISLAVSKY_KEY=sk-stan-live-…
```

tools/list

One JSON-RPC message in, one JSON body out — no session id, no event stream.

curl

curl -s https://stanislavsky.ai/api/mcp \
  -H "Authorization: Bearer $STANISLAVSKY_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

tools/call

Take a name from tools/list and pass it with the production it should read — this body is what goes after -d.

the request body

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "read_project",
    "arguments": { "project": 42, "section": "scenes" }
  }
}

5
What the server answers
initialize, ping, tools/list, tools/call. Anything else is a method-not-found, which is what a client expects from a server that declared no more.

Open your API keys

Studio staff, for now. A key is issued to an account in the studio; the team is using it first and finding which tools earn their place. Ask for access.

What you can ask

Questions a department would answer.

A few to start from: copy one into Claude once you are connected. Each names the tools it takes, and every one of those is described below.

6 examples

Overview

Where the production stands

“List my projects, open LURIA, and write me where it stands: what the story is, how many scenes and shots are broken down, what has been rendered, and what is still missing.”

list_projectsread_project

Story

Which action has no shot

“Read scene 7, then read its shot list. Which action sentences in the scene are covered by no shot at all — and for the ones that are covered, which shot covers them?”

read_sceneread_shots

Story

Where a line is said

“Search the screenplay for ‘the letter’ and tell me every scene it turns up in, with one line on what that scene is doing dramatically.”

search_scriptread_scene

Camera

Check a take before I do

“Find every take rendered for shot 7.3, tell me which one is the pick, then look at the two best frames — is the light in them what the gaffer planned for that scene?”

read_shotread_mediaview_imageread_department

Art

Check the blocking against the plan

“Read the Director's brief for the production designer on scene 4, then check what art exists for that scene: does the blocking match the floor plan, and are the props in the breakdown in the frame?”

read_briefsread_scene_artread_scene

Delivery

Why a generation failed

“Show me the last failed generation on this project: the prompt that went out, where the references came from, exactly what the provider answered, and whether it was failing for everyone that hour.”

read_generation_failureread_activity

How it behaves

It reads. It never writes.

A key is a reader, and the boundary is the account it belongs to. Everything the server does is on the record of that key.

Your key, your productions

A key opens exactly what its account can open, and no more. Revoke it and the connection stops on its next call.

Read-only by design

No edit, no render, no spend. When the answer is a change, Claude hands you the link that opens the production — the click is yours.

Answers cut to fit

One reply carries about 20,000 tokens. A longer answer is cut, and says where — narrow the question and it goes deeper.

Every call on the record

Tool, production, key, arguments, outcome and how long it took — kept with the key's own history, not in a log nobody reads.

Functions

Every tool, and what it reads.

The whole server, grouped the way a production is: what it is, what happens in it, how it is shot, what it looks like, and what went wrong.

16 read-only

The production

2 tools

Find a production, then take its shape in one answer.

list_projects

List projects

The productions this key can open, newest first, each with the link that opens it in the studio.

· id, name, workspace, status, start date
· a link back into the studio

read_project

Read a project

One production the way its crew sees it — story, structure, the cut, the cast, the breakdown and shot list, the briefs, the art, the renders, the activity. Big productions come in sections.

· a line per scene and per shot
· which sections it left out, and what to ask for

Story & scenes

4 tools

What happens, to whom, and where it is said.

read_scene

Read a scene

One scene whole: the screenplay's own action and dialogue, the 1st AD's card, the value charge and the rhythm sketch.

· verbatim action and dialogue
· presence, props, wardrobe, coverage

search_script

Search the screenplay

Every line the screenplay and the scene summaries have that matches a phrase, each with its scene number.

· matching lines, numbered by scene

read_character

Read a character

One person: their screenplay entry, the arc, every scene they are in, and the casting profile if one has been made.

· want, need, description
· the arc and every appearance
· casting profile and looks

read_briefs

Read the Director's briefs

What the Director asked each department for — task, intent, constraints and references — by department or by scene.

· per department, per scene

Camera & departments

3 tools

The shot list, one shot, and any department's own work.

read_shots

Read the shot list

The complete shot package for one scene, or the whole film: size, angle, movement, lens, composition, continuity, intent, duration.

· every field of every shot
· what each one covers, who is in frame

read_shot

Read a shot

One shot whole: its row in the list, the camera override, the lighting plan, every still and take rendered for it — and which one is the pick.

· the row and the override
· the lighting plan
· stills, takes, the pick

read_department

Read a department

What one department has on this production, in its own depth — the Gaffer's plans, Costume's looks, the DoP's list, the 1st AD's breakdown.

· one department's work, as its own

Art, media & evidence

5 tools

The plates, the documents, the renders — and the frames themselves.

read_scene_art

Read a scene's art

The Production Designer's work on one scene: the space, the plan, the blocking, the base storyboard — whichever of the four exist.

· space · plan · blocking · storyboard

list_artifacts

List artifacts

Every document the pipeline has made for the production, with its status and size — what exists before asking for it.

· code, status, last updated, size

read_artifact

Read an artifact

One document verbatim by its pipeline code: screenplay, structure, virtual edit, breakdown, shot list, briefs, or a scene's art. Large ones are cut, so prefer the focused tools.

· the raw document, by code

read_media

List rendered media

The index of what has actually been rendered — portraits, looks, art, panels, stills, clips, takes — with the model, the date and whether a take is the pick. Counts, not pixels.

· model, date, select or circled
· how many takes of which shot

view_image

Look at a frame

The image itself, not its record: composition, light, likeness, whether a take works. Claude looks at the frame.

· the actual frame, as an image

What happened

2 tools

What the crew ran, and what the studio refused.

read_activity

Read crew activity

The production's log, newest first: which agent ran what, when, on which model, for how long, at what cost, and whether it failed.

· agent, model, duration, cost
· what failed

read_generation_failure

Read a failed generation

One failure in full: the window, the model, the prompt, where each reference came from, what the provider answered, and how that model was doing for the whole studio.

· prompt and references
· the provider's own words
· the model's health that hour

Nothing here writes. No edit, no render, no spend — a tool asked for one answers with the link that opens the production in the studio, where the click is yours. Every call is written down against the key that made it.

Put Claude on the production.

Make a key, run one command, ask. Keys are for studio staff for now — the team is using it first, and everyone else can ask for the door to open.

Make a key Request access

https://stanislavsky.ai/api/mcp