MCP server for social media management through Simplified API, providing dynamic workflow tools and OpenAPI-based integrations
This MCP server exposes 21 tools for media processing (video, audio, images) via the Simplified API. While all tools have descriptions and input schemas are present in the specification, the quality is uneven. Naming is generally good (verb_noun pattern: transcribe_video, blur_background, convert_video_format). However, descriptions are inconsistent in depth and actionability. Many tools lack details about expected output format, error conditions, and when to use similar tools (e.g., distinguish between blur_background vs remove_background). Parameter descriptions exist but are often minimal (e.g., 'URL of the source image' without format constraints or examples). Critical gaps: (1) Output schemas are not documented in the code sample provided, we cannot verify what fields are returned, which breaks downstream tool chaining and LLM planning. (2) No error handling strategy is visible, tools do not document retryable vs fatal errors or recovery guidance. (3) Parameters lack constraint clarity, blur_value ranges 1-100 but there's no guidance on practical values or visual effect ranges. (4) Complex tools like script_to_video and text_to_video have deeply nested payload objects but lack sufficient documentation of required vs optional nested fields. (5) Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are completely absent despite many tools being stateful writes. Overall: definitions are present but fall short of production quality. Average per-tool score: 48/100.
Auto-add B-roll footage to a video
Blur image background
Convert video to a different format
Download transcription as WebVTT or SRT (step 3 of 3)
Generative fill / inpaint with prompt
Poll a transcription job for status and transcript (step 2 of 3)
Outpaint / extend image beyond borders
Output schemas not documented. Tools define input parameters but the code sample does not show what fields are returned. This breaks LLM planning for chained tool calls and forces agents to guess at response structure.
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint tags on any tools despite clear risk stratification (WRITE vs READ_ONLY). Agents cannot distinguish retry-safe tools from state-modifying operations.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Magic inpaint — AI object removal / replacement
Merge multiple videos into one
Pix-to-pix image transformation
Remove audio track from video
Remove image background
Replace image background
Restore / enhance image quality
Reverse a video
Generate a video from a script
Scribble to image (ControlNet)
Change video playback speed
Generate a video from a text prompt
Start a video/audio transcription job (step 1 of 3)
Upscale image resolution
Descriptions lack actionability and depth. Most descriptions are 40 - 60 characters, falling short of the 50 - 200 character baseline. Example: 'Blur image background' does not explain when to use blur_background vs remove_background, expected output format, or timeout behavior. Descriptions must be LLM-optimized.
Parameter descriptions are minimal and lack context. Examples: 'URL of the source image' does not specify supported formats (JPEG, PNG, WebP?), size limits, or timeout expectations. 'Blur intensity (1 - 100)' lacks guidance on perceptual effect (what does 50 look like vs 80?). Parameters must include format, range rationale, and practical guidance.
No error handling strategy documented. Tools do not explain which errors are retryable (transient API failure, rate limit) vs user-fixable (invalid format, unsupported language) vs fatal. Agents cannot recover from failures without explicit guidance.
Complex nested parameters (script_to_video, text_to_video) lack clarity on which nested fields are required vs optional. The payload object contains 11 properties but only 'title' is marked required. Agents cannot construct valid payloads without guessing.
Tool naming creates ambiguity. 'blur_background' vs 'remove_background' vs 'replace_image_background', the distinctions are not obvious from names alone. LLMs will struggle to pick the correct tool. Descriptions should explicitly state when to use each.
sd_scribble tool has inconsistent parameter documentation. Properties like 'image_resolution', 'steps', 'guidance_scale', 'counts' lack descriptions, and it's unclear whether 'image_url' and 'image' are alternatives or both required. Schema is ambiguous.
No pagination or result limits documented. Tools like 'get_transcription' and 'download_transcription_file' do not specify whether they return paginated results or full payloads. Large responses could exhaust context windows.