PrismClip

API and MCP integrations

Connect PrismClip to an AI-assisted video workflow.

PrismClip provides OAuth-connected MCP, owner-scoped API keys, durable jobs, and a published OpenAPI contract. Public operations ingest sources, search owned evidence, create and revise moment sets, queue automatic clip suggestions, and read durable state.

Add PrismClip to an MCP client Browse the API
01

Owner-scoped access

MCP clients connect through PrismClip OAuth. REST integrations can create a named API key with only the scopes the workflow needs.

02

Shared contracts

REST validation, OpenAPI operations, and MCP tools derive from the same Zod schemas and operation registry.

03

Durable work

Long-running operations return job IDs. Stable idempotency keys make uncertain requests safe to retry.

MCP connection

Add PrismClip with one command.

Your MCP client discovers PrismClip OAuth from the server URL and opens a browser for authorization. The same URL can be added to clients such as Claude and ChatGPT that support remote Streamable HTTP MCP servers.

codex mcp add prismclip --url https://prismclip.com/mcp

REST and headless authentication

Create a scoped API key from your account.

Send the key in the Authorization header as a Bearer token. API keys support REST automation and manually managed MCP clients. Keep keys in a secret manager and use the narrowest scopes the workflow needs.

Authorization: Bearer $PRISMCLIP_API_KEY

REST quick start

Queue suggestions for an analyzed project.

Use a project owned by the key’s account after full visual analysis completes. Reuse the same idempotency key when retrying this exact request, then poll the returned job URL.

curl https://prismclip.com/api/v1/projects/PROJECT_ID/auto-pick \
  --request POST \
  --header "Authorization: Bearer $PRISMCLIP_API_KEY" \
  --header "Idempotency-Key: workflow-2026-07-15-001" \
	  --header "Content-Type: application/json" \
	  --data '{"count": 6}'

Remote sources

Import from Google Drive or YouTube.

Paste a Google Drive or YouTube URL. Private Drive files use the account connection created through PrismClip’s Google OAuth flow. Both imports return a job that you can check or cancel.

curl https://prismclip.com/api/v1/imports/google-drive \
  --request POST \
  --header "Authorization: Bearer $PRISMCLIP_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{"url":"https://drive.google.com/file/d/FILE_ID/view"}'

curl https://prismclip.com/api/v1/imports/IMPORT_JOB_ID \
  --header "Authorization: Bearer $PRISMCLIP_API_KEY"

Transcript and visual evidence

Search owned recordings and inspect exact source timing.

Library search matches filenames, summaries, transcript words, and completed visual scenes. Transcript results include word ordinals and source-relative milliseconds. Visual scene results include the analysis ID, model, duration, timestamps, and stable scene indices used by evidence-backed moment operations.

search_library({ q: "heated audience exchange", limit: 20 })

get_transcript_passage({
  project_id: "PROJECT_ID",
  start_ms: 12000,
  end_ms: 30000,
  limit: 200
})

get_visual_scenes({ project_id: "PROJECT_ID", offset: 0, limit: 100 })

Explicit moments

Create complete moments from cited source evidence.

Pass transcript word ordinals, visual scene indices, or both with each ordered source range. PrismClip validates citations against the project’s stored evidence. Passage revisions create a new immutable moment set and use the same range, evidence, ordering, and caption-correction rules as Studio review.

create_moment_set({
  project_id: "PROJECT_ID",
  claim: "Heated audience exchange",
  target_seconds: 30,
  passages: [{
    evidence_kind: "mixed",
    first_word: 120,
    last_word: 168,
    start_ms: 42000,
    end_ms: 57000,
    scene_indices: [8, 9],
    caption_corrections: [],
    reason: "A complete exchange followed by the audience reaction."
  }]
})

get_moment_set({ project_id: "PROJECT_ID", moment_set_id: "MOMENT_SET_ID" })

Cloud rendering

Queue a durable MP4 and retrieve its output.

Cloud renders use the active owned moment set and preserve the selected passage IDs, output settings, progress, and renderer version. Completed state includes API-key-authenticated preview and download URLs. Active renders can be cancelled. Terminal renders can be deleted while no feed submission depends on them.

queue_render({
  project_id: "PROJECT_ID",
  moment_set_id: "MOMENT_SET_ID",
  aspect_ratio: "9:16",
  framing: "background-fill",
  captions: true,
  caption_style: "progressive",
  transition: "hard-cut"
})

get_render({ project_id: "PROJECT_ID", render_id: "RENDER_ID" })

Feed submissions

Send an eligible vertical clip to human review.

A completed 9:16 render can enter the normal feed review workflow after the account confirms publication rights. PrismClip captures the source transcript, evidence passages, render settings, and available analysis provenance for moderators. Integrations can read submission state and revise a pending title or category. Approval, rejection, withdrawal, and publication remain in the human control panel.

create_feed_submission({
  project_id: "PROJECT_ID",
  render_id: "RENDER_ID",
  title: "Complete stage exchange",
  category: "interviews",
  rights_confirmed: true
})

get_feed_submission({ project_id: "PROJECT_ID", submission_id: "SUBMISSION_ID" })

From source to clip

Connect automation with human review.

  1. 1
    Add and analyze

    Upload a local file, connect a Google Drive file, or import a YouTube video.

  2. 2
    Inspect and select

    Read the stored transcript and visual evidence, then request suggestions or create moment sets through REST or MCP.

  3. 3
    Review and render

    Revise passages through the integration or Studio. Studio includes source playback, framing, captions, and rendering.

  4. 4
    Submit for review

    Send clips to the human moderation workflow. The admin control panel handles approval and publication.

Shared knowledge

Guidance published to developers and MCP clients.

These versioned guides are the same sources returned by the MCP server under stable prismclip://guides/ URIs.

PrismClip workflow

Use PrismClip as the durable video-processing layer in an AI-assisted workflow. Work with projects owned by the API-key principal. Wait for asynchronous jobs to reach a terminal state before starting dependent work. Use transcript or visual evidence to create complete moments, render those moments, and keep a person responsible for review and publication decisions.

The Studio and Library remain complete interfaces for the same projects. API and MCP operations add integration access; they do not transfer ownership or moderation authority.

Choose a source

Upload the original file directly.

Add a Google Drive file from the connected account. PrismClip creates a cloud copy.

Add a YouTube video.

Check the import job until ingestion completes.

Evidence and moments

Use transcript evidence for spoken answers, arguments, explanations, jokes, and stories. Use visual evidence for actions or scenes that are not adequately described by speech. Use mixed evidence when both the words and visible event define the moment.

Search the owned library to locate candidate projects, then retrieve bounded transcript passages or visual scene pages. Transcript words include ordinals and source-relative milliseconds. Visual scenes include stable indices plus analysis ID, model, duration, and timestamps. Retain those identifiers when creating or revising a moment.

Keep source ranges inside the cited evidence. Create complete exchanges, actions, and thoughts. Preserve evidence references so a person can review each selection in context.

Jobs and idempotency

Long-running PrismClip operations return durable job IDs. Poll at a modest interval and stop when the job reaches a terminal state. Do not create dependent work while a prerequisite job remains queued or running.

Use one stable Idempotency-Key for retries of the same intended operation. Use a new key when the intended input changes. A replay can return the original job instead of creating duplicate work.

Rendering and review

Choose aspect ratio, framing, captions, and source ranges deliberately. Review the opening frame, ending boundary, crop behavior, caption accuracy, and source context before using a completed clip.

Use create_feed_submission after a completed 9:16 render and confirm that the account has permission to publish the clip and its contents. The submission captures durable moderation provenance. Use get_feed_submission to read state and revise_feed_submission to replace the title or category while the submission remains pending. Public API keys and MCP tools never approve, reject, withdraw, or publish feed submissions.

Errors and recovery

Correct validation and ownership errors before retrying. Respect Retry-After on rate limits. For provider failures, preserve the reported category and let the user choose another source path when access or availability is the cause.

Reuse the same idempotency key after transport interruption or an uncertain response. Create a new key only for a changed request. Include PrismClip request and job IDs in support reports without including API keys or source credentials.

Inspect every live operation.

The API reference is generated from the production contract and includes scopes, request fields, responses, errors, and MCP mappings.

Open API reference