This server provides focused access to Luma's video modification API through RunAPI, enabling AI agents to create, monitor, and price video tasks.
Authenticate (login): Log in via a browser-based PKCE flow to save credentials to ~/.config/runapi/config.json. Supports a force option to re-authenticate even when valid credentials exist. Alternatively, set the RUNAPI_API_KEY environment variable for headless/CI use.
Create Video Modification Tasks (modify_video): Submit a Luma video modification task with a source video URL and prompt. Supports optional polling until completion (wait: true/false), configurable timeout, and poll interval. Returns a task ID, status, output URLs, and a pricing snapshot.
Poll Task Status (get_task): Fetch the current status and latest result payload for an existing Luma task by its task ID — useful for checking on tasks submitted without waiting.
Check Pricing (check_pricing): Look up the current pricing snapshot for the Luma model and endpoint. No authentication required.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@@runapi.ai/luma-mcp modify video at https://example.com/vid.mp4 to apply a cinematic look"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
@runapi.ai/luma-mcp is a focused Model Context Protocol server for the Luma model line on RunAPI.
It gives MCP-compatible assistants direct access to 1 endpoint and 1 model variant without loading the full RunAPI catalog.
Use this per-model server when an agent should stay scoped to Luma. Use @runapi.ai/mcp when one assistant should discover every RunAPI model line.
Related MCP server: kling-mcp
Add it to Claude Code:
claude mcp add luma -s user -- npx -y @runapi.ai/luma-mcpUse project scope when the server should be shared with a repository:
claude mcp add luma -s project -- npx -y @runapi.ai/luma-mcpCodex, Cursor, Windsurf, VS Code, Roo Code, and other MCP hosts can use the same stdio command:
{
"mcpServers": {
"luma": {
"command": "npx",
"args": ["-y", "@runapi.ai/luma-mcp"]
}
}
}check_pricing works before sign-in. For task creation and status polling, ask your assistant to call the login tool. It opens a browser login and saves credentials to ~/.config/runapi/config.json, the same file used by runapi login.
Headless and CI hosts can still set RUNAPI_API_KEY before starting the MCP host.
Ready-made examples are in examples/ for Claude, Cursor, Windsurf, VS Code, and Roo Code.
Tool | Auth | Purpose |
| Yes | Create a Luma modify video task and optionally wait for a terminal status. Returns the task id, status, and output URLs. |
| Yes | Fetch the current status and latest payload for an existing task. |
| No | Look up current pricing for a Luma model and endpoint. |
Luma covers 1 model variant across 1 endpoint. Each tool accepts the models listed for it:
Tool | Models |
|
|
Model availability can change between releases. Use check_pricing or the Luma model page for the current catalog view.
Ask your assistant in natural language; it can inspect pricing, create the task, and return the task id plus output URLs.
Run a Luma modify video task with RunAPI.The assistant can call check_pricing, then modify_video, and return the task id, status, and output URLs.
Create the task but don't wait for it to finish.The assistant calls the create tool with wait: false and returns the task id. Check on it later with get_task.
Check current Luma pricing, then create the task if it matches my request.The assistant calls check_pricing and can link to the Luma model page for the canonical catalog entry.
The server resolves auth in this order:
RUNAPI_API_KEY environment variable, useful for headless and CI hosts
~/.config/runapi/config.json, created by the MCP login tool or runapi login
No key, which still allows check_pricing
The config file is normally managed by login. A pre-provisioned headless config can use:
{
"apiKey": "your_runapi_key"
}Do not commit real API keys.
Resource | URL |
Luma model page | |
npm package | |
GitHub repository | |
RunAPI MCP overview | |
RunAPI docs |
Licensed under the Apache License, Version 2.0.
Look up RunAPI pricing for the luma model line.
| Name | Required | Description | Default |
|---|---|---|---|
| model | No | Model slug. Defaults to the line's primary model. | |
| action | No | Endpoint name. Defaults to the endpoint that offers the model. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden. 'Look up' implies a read, but nothing states auth requirements, whether pricing is cached/live, or what data comes back; for a zero-annotation tool this is a notable gap.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single tight sentence with the resource and scope front-loaded and no filler. It is efficient, though the extreme brevity is part of why behavioral context is missing.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple two-optional-param lookup with 100% schema coverage and no output schema, the definition is minimally workable. However, with no annotations it should have described the return shape (e.g., price fields/currency) and any auth or scope constraints.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the two optional parameters (model, action) and their defaults are already documented. The description only reinforces the 'luma model line' default context, adding little beyond the schema, making the baseline 3 appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Clear verb+resource+scope: 'Look up RunAPI pricing for the luma model line' tells an agent exactly what it retrieves and for which model family. It is distinguishable from siblings like login/get_task/modify_video, though it never names or contrasts them explicitly.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Provides no when-to-use guidance, prerequisites, or alternatives. An agent is not told when this should be called relative to modify_video or get_task, nor whether it needs to log in first.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Fetch the current status and latest result payload for a luma task.
| Name | Required | Description | Default |
|---|---|---|---|
| action | No | Asynchronous endpoint the task was created on. Defaults to the line's only asynchronous endpoint. | |
| task_id | Yes | Task id returned when the task was created. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden. It does disclose read semantics and that the payload returned is the 'latest result', implying results may not exist yet. But it omits auth requirements, whether repeated calls are safe, and how status values should be interpreted.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single tight sentence that front-loads the verb and resource with no wasted words. Nothing redundant or padded.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema and no annotations, so the description is the only place an agent could learn what the returned status/result payload looks like or when polling should stop. For a status-polling tool it leaves the workflow and return shape underspecified.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so both parameters (task_id, action) are already documented in the schema, including the enum and its default behavior. The description adds no parameter-level meaning, so the baseline 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb (Fetch) and resource (luma task status + latest result payload), which is unambiguous. However it does not differentiate itself from any sibling — the siblings (login, modify_video, check_pricing) don't overlap, so there is nothing to contrast against, which keeps it short of a 5.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No when-to-use guidance is given: nothing says this is the polling tool to call after modify_video returns a task id, nor whether it should be called once or repeatedly until a terminal state. The agent must infer the workflow entirely.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Authenticate RunAPI by opening a browser PKCE login flow and saving the API key to ~/.config/runapi/config.json.
| Name | Required | Description | Default |
|---|---|---|---|
| force | No | Re-run browser login when the current credential comes from the local config file. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the behavioral burden. It discloses the interactive browser flow and the file write side effect (config.json). However, it does not mention that it may overwrite existing credentials or that it could block waiting for user input, though these are implied.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, well-structured sentence that front-loads the action ('Authenticate RunAPI') and provides necessary details without extraneous information.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple login tool with one optional parameter and no output schema, the description covers the core purpose and side effect. It lacks an explicit statement that this is a prerequisite for other tools, but that is implied.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% (the only parameter 'force' has a description). The tool description adds no additional meaning about parameters beyond the schema, so the baseline of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's purpose with a specific verb ('Authenticate'), target resource ('RunAPI'), method ('browser PKCE login flow'), and side effect (saving to config.json). It is distinct from sibling tools, none of which relate to authentication.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage (to authenticate RunAPI) but does not explicitly say when to run it (e.g., before other RunAPI tools) or when to use the 'force' parameter. Since there are no alternative auth tools among siblings, 'vs alternatives' is not applicable.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Create a Luma task on RunAPI (modify video). Returns a task id, status, and output URLs.
| Name | Required | Description | Default |
|---|---|---|---|
| wait | No | Poll until the task reaches a terminal status. | |
| model | No | RunAPI model slug for this model line. | |
| prompt | Yes | Declared type: string. | |
| watermark | No | Declared type: string. | |
| timeout_ms | No | ||
| callback_url | No | Declared type: string. | |
| poll_interval_ms | No | ||
| source_video_url | Yes | Declared type: string. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the full burden, and it does disclose the return shape (task id, status, output URLs), which matters since no output schema exists. However, it omits async/long-running behavior, callback vs polling semantics, cost implications, and whether the task is reversible or consumptive.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two tight sentences with the action front-loaded and the return values appended. Nothing is wasted, though the content is thin rather than merely concise.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For an async task-creation tool with 8 parameters, no annotations, and no output schema, the description partially compensates by naming the return fields. It still leaves authentication, cost, model selection, and polling behavior unexplained, which is a meaningful gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The description adds no meaning for any of the 8 parameters. Schema coverage of 75% is inflated by placeholder descriptions like 'Declared type: string', so watermark, callback_url, model, timeout_ms, and poll_interval_ms are effectively undocumented in both places.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a verb+resource (create a Luma task for modifying video) and names the backend, but 'modify video' is vague about what transformation actually occurs. It does not distinguish itself from the sibling get_task, which is the natural follow-up for the returned task id.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
There is no guidance on when to call this versus get_task, check_pricing, or login, nor on prerequisites such as needing an authenticated session. The `wait` parameter implies synchronous polling but the description never explains when to set it.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.2.0check_pricing2 fields changedInput schema / additionalPropertiesRemoved value: -falseInput schema / properties / model / enumRemoved value: -[
- "luma-modify-video"
-]get_task1 field changedInput schema / additionalPropertiesRemoved value: -falsemodify_video8 fields changedInput schema / additionalPropertiesPrevious value: -falseNew value: +{}Input schema / properties / callback_url / descriptionAdded value: +"Declared type: string."Input schema / properties / model / enumRemoved value: -[
- "luma-modify-video"
-]Input schema / properties / poll_interval_ms / maximumAdded value: +9007199254740991Input schema / properties / prompt / descriptionAdded value: +"Declared type: string."Input schema / properties / source_video_url / descriptionAdded value: +"Declared type: string."Input schema / properties / timeout_ms / maximumAdded value: +9007199254740991Input schema / properties / watermark / descriptionAdded value: +"Declared type: string."v0.1.7get_task1 field changedInput schema / properties / action / descriptionPrevious value: -"Endpoint the task was created on. Defaults to the line's only endpoint."New value: +"Asynchronous endpoint the task was created on. Defaults to the line's only asynchronous endpoint."modify_video5 fields changedInput schema / properties / callback_urlAdded value: +{
+ "type": "string"
+}Input schema / properties / prompt / typeAdded value: +"string"Input schema / properties / source_video_url / typeAdded value: +"string"Input schema / properties / watermarkAdded value: +{
+ "type": "string"
+}Input schema / requiredAdded value: +[
+ "prompt",
+ "source_video_url"
+]v0.1.6loginv0.1.4check_pricingget_taskmodify_videoScored across 4 tools
Each tool targets a clearly distinct concern: authentication (login), task creation (modify_video), task polling (get_task), and pricing lookup (check_pricing). There is no overlap in purpose, so an agent can select correctly without hesitation.
Three of four tools follow a clean verb_noun snake_case pattern (modify_video, get_task, check_pricing). The lone outlier 'login' uses a bare verb rather than verb_object, which is a minor but forgivable deviation.
Four tools is reasonable for a narrowly scoped wrapper around a single Luma model line, and each tool serves a purpose. It leans slightly thin, since only one tool actually performs the core operation.
The surface covers auth, task creation, status polling, and pricing, but omits obvious lifecycle operations such as canceling a task, listing tasks, or other Luma video operations (e.g. generate/extend). Agents hitting these needs would be stuck with no tool to call.
Create and manage cinematic AI video renders through the Future Video Studio Agent API.
LLM chat, text tools, image generation, editing, batch image jobs, and asynchronous video generation
Create AI images, video, and audio, quote generation costs, and edit Sequencer projects.
Run multi-step AI pipelines for video, image, audio and text: upload media, run, poll results.
