MCP App Store
by KeenEthics
Nim icon

Nim MCP Server for ChatGPT

By Nim AI Inc.

Overview

Nim generates images and videos from ChatGPT using a range of AI models. Prompt-writing skills help turn a short request into a detailed, model-specific prompt. Browse templates for common creative tasks and upload your own photos and clips as references. Tools cover lip sync and image and video upscaling. Follow each generation as it runs and check your credit balance in the chat. Generations use credits from your Nim account, and results vary by model and prompt.

How to connect

  1. ChatGPT: Open the Nim listing in the ChatGPT app directory and choose Connect. ChatGPT listing Nim exposes 21 tools here.

Tools

describe_video

ChatGPT
Generate a text description/prompt from an uploaded Nim video. Call this tool once with no files to receive a validation error listing the current required/optional inputs (video file, and any prompt/field the graph currently exposes), then call again with those inputs filled in via files/prompts/fields. Local files and Claude attachments must be uploaded with media_upload first, then the returned file_url should be passed under files. This tool returns immediately with status="queued"; poll get_generation_status with the returned workflowId/promptId until it reaches a terminal status, then read outputText for the result -- that polled workflow also reports mediaUrl, but for this tool that is just an echo of the source video, not the description. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

explore_templates

ChatGPT
Browse Nim templates and show selectable template cards. Call this ONLY when the user explicitly asks about templates — e.g. they use the word "template", ask to find/browse Nim templates, or ask to run a specific named template. Do NOT call it for ordinary image or video requests; use generate_image/generate_video for those. After the user picks a template, call get_template for its templateContract, collect the required inputs, upload local or attached reference files with media_upload, then call run_template. Never create HTML artifacts or local runner apps instead of run_template.

generate_image

ChatGPT
Start an async Nim image generation. Call models_explore with action=get for the chosen model first, then follow generationContract exactly. This tool returns no result and no link; the generation runs asynchronously. If the attached Nim generation widget actually renders (MCP-Apps hosts such as claude.ai, ChatGPT, or Claude Desktop), let it display progress and the final media. Otherwise — including terminal/CLI clients such as Claude Code and Codex — you MUST poll get_generation_status with the returned workflowId/promptId until it reports a terminal status (finished, failed, cancelled, or removed), then report the result. The result includes estimatedDurationMs — the typical processing time for this model measured from recent generations (excludes queue wait); pace your polling with it as suggested in the result text, and treat generations running past the estimate as normal. Never tell the user the result will appear in a widget unless one is actually shown, and never invent external Nim links. Never embed mediaUrl as an inline markdown image — many hosts block external images and render a broken placeholder; when reporting a result without a visible widget, share mediaUrl as a plain URL. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

generate_video

ChatGPT
Start an async Nim video generation. Call models_explore with action=get for the chosen model first, then follow generationContract exactly. This tool returns no result and no link; the generation runs asynchronously. If the attached Nim generation widget actually renders (MCP-Apps hosts such as claude.ai, ChatGPT, or Claude Desktop), let it display progress and the final media. Otherwise — including terminal/CLI clients such as Claude Code and Codex — you MUST poll get_generation_status with the returned workflowId/promptId until it reports a terminal status (finished, failed, cancelled, or removed), then report the result. The result includes estimatedDurationMs — the typical processing time for this model measured from recent generations (excludes queue wait); pace your polling with it as suggested in the result text, and treat generations running past the estimate as normal. Never tell the user the result will appear in a widget unless one is actually shown, and never invent external Nim links. Never embed mediaUrl as an inline markdown image — many hosts block external images and render a broken placeholder; when reporting a result without a visible widget, share mediaUrl as a plain URL. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

generate_voice_sample

ChatGPT
Generate three voice sample variants from a character description with ElevenLabs Voice Design — e.g. "grandpa fisherman voice", "grumpy wizard", "upbeat Gen-Z podcaster". Costs 6 Nim credits per call for all three variants; say so before calling. You write both inputs in English: voice_description (60–100 words: age, gender, timbre, pace, accent, emotion, who the character is) and text (100–200 characters, in character, no double quotes). Tell the user which voice parameters you chose. The result has three mp3 URLs. If a Nim voice samples widget renders in this conversation, it plays them inline — do not repeat the links under it. Otherwise give the user the links (they open in any browser) or download them with curl into the working folder. If no variant fits, ask what to change (lower, slower, angrier…) and call again with an adjusted voice_description; every call is a new charge. Nothing is saved to a voice library; the samples are disposable previews. If the user lacks credits, this returns status="insufficient_credits" instead of generating — show its message to the user and do not retry.

get_credit_balance

ChatGPT
Returns the authenticated Nim account email and current effective credit balance, including subscription, creator program, pack, grant, and total available credits. Use before expensive generation requests or to identify the connected account.

get_generation_status

ChatGPT
Poll the status of an async Nim image/video generation by workflowId or promptId until it reaches a terminal status (finished, failed, cancelled, or removed). After generate_image or generate_video, call this repeatedly to track progress and obtain the final media URL — this is required for terminal/CLI clients (e.g. Claude Code, Codex), which do not render the Nim generation widget. Non-terminal responses include estimatedDurationMs (typical processing time for this model, excludes queue wait) and startedAt — pace polling with the suggested interval in the result text, and treat generations running past the estimate as normal, not failed. Exception: when the Nim generation widget is actually rendered (MCP-Apps hosts), do not narrate polling or show links; let the widget display progress and the final media. When you report a finished result yourself, present mediaUrl as a plain URL — never as an inline markdown image, because many hosts block external images and render a broken placeholder. It also returns another user's generation that this user can open on nim.video (pass the promptId from its nim.video/explore/<promptId> link): its media and prompt, watermarked as nim.video shows them to this user, without the owner's project, cost or error.

get_template

ChatGPT
Inspect a single Nim template selected with explore_templates and return its templateContract (required and optional inputs). Collect every required input from the user, and ask about optional user-facing settings when they materially change the result, especially Number of images/videos, Photo type, and Aspect ratio. Use templateContract labels when talking to the user; use names only as run_template input keys. Upload local or attached reference files with media_upload, then call run_template with namespaced inputs. Never create HTML artifacts or local runner apps instead of run_template.

lipsync

ChatGPT
Animate the face in a Nim-hosted image or video to speak: sync its lips to uploaded speech audio (audio_url) or to synthesized speech from text (speech_text + voice). The output is always a video. Accepts the file_url from media_upload or the mediaUrl of a finished generation from get_generation_status; for external files call media_upload first. Pricing is per second of speech — check estimatedCreditCost in the result. Requires a one-time user consent: if the result has status="consent_required", show the user the consentUrl and ask them to accept it in the browser themselves (never claim consent on their behalf), then call this tool again. The job runs asynchronously: poll get_generation_status with the returned workflowId/promptId until a terminal status, unless the attached Nim generation widget actually renders (then let it display progress). If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

list_projects

ChatGPT
List the Nim projects, with their folders, the user can save generations into: their own projects, workspace (team) projects, and projects shared with them for editing, most recently active first. Call it only when the user names a project or folder or asks where results are saved, not before every generation: without project_id, generation tools save to the default project. Pass a returned project_id as project_id to generate_image, generate_video, run_template, upscale_image, upscale_video, lipsync, describe_video, or render_video_draft. Each project lists its folders (folder_id, name, parent_folder_id and a "Parent / Child" path, sorted by path); pass a folder_id as folder_id to those tools only when the user names a folder, otherwise results go to the project root. If several folders match what the user said, ask which one they mean. foldersTruncated means the project has more folders than listed: narrow the projects with query or ask the user. query is a plain case-insensitive substring match: if it finds nothing (typo, other language, partial words), call again without query and match the names yourself, or ask the user. If several projects match what the user said, ask which one they mean. isDefault marks where generations go without project_id. mode PRIVATE means only the owner can see the project; other modes share it with other people. This tool never creates projects.

media_upload

ChatGPT
Upload a local file or Claude attachment to Nim before using it as reference media. Call this tool whenever another Nim tool needs fileInputs and the available file is a local path or Claude attachment path such as /mnt/user-data/uploads/file.jpg. After this tool returns curl_example, run it with the actual local file path, then use the response file_url in generate_image, generate_video, or run_template. Do not pass local filesystem paths directly to generation tools. No Authorization header is needed — authentication is embedded in the upload URL. Supported formats: image/jpeg, image/png, image/webp, image/gif, video/mp4, video/quicktime, audio/mpeg, audio/wav, audio/x-wav, audio/mp4. Max 20 MB (see forTool for a higher limit). The curl response JSON contains file_id and file_url for use in other Nim tools.

models_explore

ChatGPT
Find Nim generation models. Use list/search/recommend capability summaries to compare models. Call get only for the final chosen model before generate_image or generate_video; get returns the exact generationContract.

open_nim_library

ChatGPT
Open the Nim media library. Browse your own image/video generations, optionally filtered by project_id from list_projects. Omit project_id for all projects. Search scans a bounded recent-history window; use nextCursor as after with the same project_id to continue.

render_video_draft

ChatGPT
Render a finished video draft (made with generate_video draft=true) in high resolution. Pass the draft promptId; the render reuses the draft prompt, references and settings, so it takes no prompt or model. Call it when get_generation_status reports canRenderDraft. The render is a new async generation: track it exactly like generate_video (the Nim generation widget, or get_generation_status polling for terminal/CLI clients). By default it goes to the project of the draft. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

run_template

ChatGPT
Start a Nim template workflow selected with explore_templates. Before calling this tool, call get_template and collect every required templateContract input from the user. Ask about optional templateContract settings that materially change the result, especially Number of images/videos, Photo type, and Aspect ratio, unless the user already specified them. Use the same reference upload flow as generate_image/generate_video: local files and Claude attachments must be uploaded with media_upload first, then the returned file_url should be passed in inputs.files. Never create HTML artifacts or local template runner apps instead of this tool. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

search_voices

ChatGPT
Find ElevenLabs voices for text_to_speech. Searches the ElevenLabs voice library (Nim staff also get the Nim team voices, usable with voice_changer too) and returns each voice's voiceId and name: pass them as voice_id and voice_name. Use it whenever the user describes a voice instead of giving an id; share the preview links if they want to choose. A generatedVoiceId from generate_voice_sample is a temporary preview, not a voice_id. Free: it does not render audio.

text_to_speech

ChatGPT
Speak text with an ElevenLabs voice (Eleven v3) and save the mp3 as a Nim generation. Use it for full voice-overs: one call reads up to 5000 characters in one consistent voice. Write the delivery into the text with audio tags, e.g. "[excited] We did it!" — turn the user's description of the emotion into tags yourself. Priced by the length of the text; the charge is in creditsCharged. The job runs asynchronously: poll get_generation_status with the returned promptId until it finishes; the mp3 is its mediaUrl. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message and do not retry. Voice generations have a daily limit per account, and a refusal for it says when the limit resets. If the account has used the limit, tell the user and do not retry before then. If the refusal says only this request is larger than what is left, do not repeat it before the reset, but a shorter text may still fit now.

track_skill_activation

ChatGPT
Records that a Nim skill is being applied, so Nim can measure how often each skill is used. Sends only the skill name: no prompt, files, or conversation content. Call it once per new user request for each Nim skill whose SKILL.md you actually loaded and follow for that request; never for a skill you only saw listed by name or description. Do not call it again while polling, retrying, or continuing the same request. The result is an acknowledgement only. If the call fails, continue the task without retrying it and without mentioning it to the user.

upscale_image

ChatGPT
Upscale a Nim-hosted image by a 2-10x factor. Accepts the file_url from media_upload or the mediaUrl of a finished generation from get_generation_status — Nim-hosted media never needs re-uploading; for external files call media_upload first. For videos use upscale_video. The job runs asynchronously: poll get_generation_status with the returned workflowId/promptId until a terminal status, unless the attached Nim generation widget actually renders (then let it display progress). If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

upscale_video

ChatGPT
Upscale a Nim-hosted video by a 2x/4x factor (or to a target resolution with the bytedance model), optionally raising the frame rate to 60fps. Accepts the file_url from media_upload or the mediaUrl of a finished generation from get_generation_status — Nim-hosted media never needs re-uploading; for external files call media_upload first. For images use upscale_image. Pricing is per second of video — check estimatedCreditCost in the result. The job runs asynchronously: poll get_generation_status with the returned workflowId/promptId until a terminal status, unless the attached Nim generation widget actually renders (then let it display progress). If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message to the user and do not retry.

voice_changer

ChatGPT
Re-voice speech from a Nim-hosted video or audio with an ElevenLabs voice and save the result as an mp3 generation. The timing and intonation of the source are kept; only the voice changes, and the output is always audio. Up to 5 minutes of source. Priced by the length of the source; the charge is in creditsCharged. Upload local files with media_upload first. The job runs asynchronously: poll get_generation_status with the returned promptId until it finishes; the mp3 is its mediaUrl. If the user lacks credits, this returns status="insufficient_credits" instead of starting a job — show its message and do not retry. Voice generations have a daily limit per account, and a refusal for it says when the limit resets. If the account has used the limit, tell the user and do not retry before then. If the refusal says only this request is larger than what is left, do not repeat it before the reset, but a shorter source may still fit now.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/nim"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/nim)