google-flow-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| FLOW_CDP_URL | No | Attach to your running Chrome instead of launching one | |
| FLOW_HEADLESS | No | Launch headless (read-only use) | 0 |
| FLOW_OUTPUT_DIR | No | Where downloads and scene exports go | ~/FlowExports |
| FLOW_CHROME_PATH | No | Which browser to launch | chrome |
| FLOW_PROFILE_DIR | No | Chrome profile for option B | ~/.google-flow-mcp/profile |
| FLOW_BROWSER_CHANNEL | No | Which browser to launch | chrome |
| FLOW_MAX_CREDITS_PER_CALL | No | Default credit cap for each generating call | 20 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| flow_statusA | Check that the MCP browser is signed in to Flow and return the remaining credit balance. |
| flow_list_modelsB | Video models with key, family, mode (text/references/start_frame/start_end_frames/edit/extend/upsample), duration, resolution and credit cost (highest tier price). |
| flow_list_voicesC | Voices that can be assigned to a character. |
| flow_list_projectsB | List the user's Flow projects. |
| flow_create_projectC | Create a new, empty Flow project. |
| flow_get_projectB | Characters, generations and media, scenes with their clips, agent instructions and generation defaults. |
| flow_create_characterB | Create a reusable character. With portrait_prompt, Flow's character builder generates its look (Nano Banana, image credits). Also sets voice and personality notes the Flow agent uses. |
| flow_update_characterC | Change a character's name, voice or personality notes. |
| flow_regenerate_portraitC | Change an existing character's portrait with an instruction ("give her a red scarf"), in Flow's character editor. |
| flow_generate_imageC | Generate images (Nano Banana) through Flow's composer, optionally with characters and reference images as ingredients. |
| flow_generate_videoA | Generate video clips through Flow's composer. character_ids / image_ids become ingredients (characters stay consistent). Choose a model by model_key or by family + duration_sec + resolution (Omni supports 360p/720p and 4-10s). SPENDS CREDITS: Flow's own quote is checked against max_credits first; dry_run=true returns the quote only. |
| flow_edit_videoB | Edit an existing clip with an instruction. Flow uses an Omni edit model (costlier than generation). SPENDS CREDITS; quote checked against max_credits. |
| flow_video_statusB | Check (or wait for) video jobs. Succeeded items include a signed video URL (valid a few hours). |
| flow_get_mediaD | An image or video with a fresh signed URL. |
| flow_download_mediaC | Download an image or video to a local file. Returns the path. |
| flow_create_sceneC | Create an empty scene (timeline) in a project. |
| flow_add_clips_to_sceneB | Place videos on a scene timeline, in order. Accepts media ids or generation (workflow) ids. Appends by default. |
| flow_export_sceneA | Combine a scene's clips into one MP4 on this machine (downloads each clip; ffmpeg normalises size, fps and audio, then joins). Returns the path. |
| flow_agent_chatA | Send a message to Flow's creative agent (brainstorms, writes prompts, generates). character_ids are attached as mentions. If it asks to spend credits it is approved only when the quoted cost <= max_credits (default 0 = never); otherwise it stays pending for flow_agent_respond. |
| flow_agent_respondC | Approve or reject the agent's pending request to spend credits. |
| flow_list_agent_sessionsC | Flow agent chat sessions in a project. |
| flow_set_agent_instructionsB | Project "Agent instructions" (guidelines Flow's agent follows: style, format, tone). mode "add" appends, "replace" sets the list, "remove" deletes by id or title. |
| flow_set_generation_defaultsC | Model defaults the agent uses (image aspect/count/model, video model/aspect/count) and whether it must ask before spending credits. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 23 tools
The tools largely target distinct resources and actions: generation vs editing vs status vs media retrieval are clearly differentiated by descriptions. There is minor potential overlap between agent-driven generation (flow_agent_chat) and direct generation (flow_generate_video/image), but the agent interface is presented as distinct. Overall, an agent should be able to select the right tool without much confusion.
All tools use a consistent flow_ prefix and snake_case, with a clear verb_noun pattern for most tools (list_models, create_project, generate_video). Slight deviations like flow_status and flow_video_status (noun-first) and the agent tools (flow_agent_respond, flow_agent_chat) break the verb_noun pattern, but the naming remains predictable and readable.
23 tools is on the heavy side for a Flow client, covering many aspects of project, character, scene, media, and agent management. While each tool addresses a distinct need, the count falls in the borderline heavy range (16-25), which may increase cognitive load for agents. A slightly leaner surface would likely be more effective.
Core workflows for projects, characters, scenes, media, generation, and agent interaction are well covered. However, deletion operations (project, character, scene) and project update are missing, and flow_list_models only lists video models, leaving minor gaps that agents can likely work around.