Frame.io MCP | Frame.io API Documentation

The Frame.io MCP, hosted at https://mcp.frame.io/v1/mcp, connects MCP-compatible clients such as Claude, ChatGPT, Cursor, VS Code, and GitHub Copilot to your Frame.io account. Once connected, you can drive your Frame.io activity with an AI agent.

This guide covers how the MCP works, how to connect to it, and what it can do once running, along with a tool reference and example prompts to get started. If you are new to MCP, the next section explains the protocol before the connection steps.

Model Context Protocol

MCP is an open standard, created by Anthropic, for connecting AI clients to external systems. A connected system exposes a set of tools, which are operations the AI agent can call. When you make a request, the agent picks the right tool, calls the server, and uses the response to resolve the request.

For more on MCP, see Anthropic’s original announcement.

How It Works

Architecture

┌───────────────┐        ┌─────────────────────┐        ┌───────────────┐
│   AI Agent    │        │     Frame.io MCP    │        │ Frame.io API  │
│(Claude, etc.) │◀──────▶│  mcp.frame.io/mcp   │◀──────▶│ api.frame.io  │
└───────────────┘        └──────────┬──────────┘        └───────────────┘
                                    │
                           ┌────────┴────────┐
                           │   Adobe Login   │
                           │  (OAuth 2.0)    │
                           └─────────────────┘

Authentication

Authentication uses OAuth 2.0 through Adobe IMS, and requires an Adobe ID associated with your Frame.io account. The MCP does not support Frame.io Legacy accounts.

Each client has its own unique connection steps, covered below, but this is essentially what happens: authentication runs once, on first connection, and after that the MCP handles token refresh automatically. You are redirected to Adobe’s login screen to sign in and authorize access, then returned to your AI agent with the connection active.

Tool calls run with your Frame.io permissions. The MCP can only do what your account role already allows, so a tool that tries to exceed your access returns a permission error rather than performing the action. See Managing user permissions for how roles map to capabilities.

Adobe Managed Accounts (AMA) and Frame Managed Accounts (FMA)

Frame.io has two types of user accounts that are administered differently, which matters during authentication. See Frame Managed vs. Adobe Managed Accounts for a full breakdown.

Frame Managed Accounts (FMA) are created and managed directly within Frame.io. Sign in with your Adobe credentials at the IMS prompt, authorize access, and you are connected.

Adobe Managed Accounts (AMA) are administered through the Adobe Admin Console, typically for enterprise teams. Authentication goes through IMS as with FMA, but the key difference is that the same Adobe identity can belong to more than one organization, sometimes labeled as an account or account profile. Whichever one you authenticate into determines which Frame.io account the MCP can access.

When the OAuth flow opens in your browser, Adobe IMS checks whether you are already logged in. If you are, it reuses that session without asking you to select a profile. If that session belongs to a different organization than the one tied to your Frame.io account, you will authenticate successfully but land in the wrong place. The MCP connects without errors, but Frame.io API calls return empty results or permission errors.

If you are on an AMA and see this behavior after connecting, see AMA multi-account authentication errors in Troubleshooting.

Connecting requires an active Frame.io account with an associated Adobe ID (more on connecting to Adobe authentication can be found here).

If these steps look different

The clients covered below change their MCP setup flows frequently. This guide is updated as those changes roll out, but if the options or labels you see differ from the steps below, check the linked client documentation in each section for the latest instructions.

Claude

ChatGPT

VS Code

Cursor

GitHub Copilot CLI

Connecting to Claude

These steps cover Claude Desktop. Claude is also available on Web, iOS, Cowork, and Code Desktop. See Anthropic’s connector documentation to confirm setup steps for the surface you are using.

Open Connector

In Claude Desktop, open Customize from the menu bar. Select Connectors in the left sidebar, then click + in the top-right corner of the Connectors panel.

Enter the server details

In the Add custom connector dialog, fill in both fields:

  Name: Frame.io
  URL: https://mcp.frame.io/v1/mcp

Click Add.

Authenticate

Frame.io appears in the Not connected section with a CUSTOM label. Click Connect in the right panel. You will be taken to Adobe IMS. Sign in with the Adobe ID associated with your Frame.io account. After authorizing, you are returned to Claude Desktop.

Review tool permissions

Once authenticated, Frame.io moves to the Web section. The right panel lists tools split into Interactive tools and Read-only tools.

Each group has a permission setting, set to Needs approval by default. Click the dropdown to change it:

Setting	Behavior
Always allow	Claude calls the tool without prompting
Needs approval	Claude pauses and asks before each tool call
Blocked	Claude cannot use these tools
Custom	Configure permissions per tool

Read-only tools are safe to set to Always allow. Leave all other tools on Needs approval until you have reviewed the Tool Approval Settings section below.

See Claude’s documentation on connector permissions for more detail.

Use Frame.io tools

Once connected, ask Claude to perform Frame.io tasks in plain language. Claude surfaces the connector’s tools automatically and picks the right one for each request.

To confirm the connection is working, ask Claude: “List my Frame.io accounts.”

Available Tools

Reference for the tools available in the Frame.io MCP. If a tool or capability you need is missing, let us know.

A Note on Naming

This MCP consolidates file, folder, and version stack listing into read_assets . Metadata values are read inline on file and folder objects rather than through a dedicated read tool.

Available tools are categorized as Setup, Read, Write, Delete, or Feedback.

Read tools retrieve data without making changes. Write tools create, update, move, or restore content. Delete tools remove content. Deleted files and folders can be restored, but deletions of other resource types are permanent.

Each tool’s category is listed alongside it below.

Browse Tools: Setup and navigation · Assets and structure · Comments · Metadata · Sharing · Membership · Uploads · Deletion and diagnostics

Tool Approval Settings

Connecting any MCP server, not just this one, lets your AI agent take actions with your access and credentials. Treat it with the same care as any integration that can reach your account or files.

Before you change any permission settings from their defaults, review how tool execution works.

When a tool category is set to Always allow, your AI agent calls those tools automatically, with no confirmation and no preview of what is about to happen. This is fine for read-only lookups. But it means a prompt like “organize these into the right folders” could move files to the wrong place before you get a chance to review where they landed.

Recommended defaults:

Tool type	Recommended handling
Read	Safe to auto-approve, if your client supports it
Write	Review each call before it runs
Delete	Review each call before it runs, these remove content

Read-only tools

Read-only tools carry the standard readOnlyHint annotation, which Claude, ChatGPT, VS Code, and Cursor use to identify reads as non-modifying, usually grouping them separately or skipping the approval prompt. The exact handling varies by client.

Every client implements approval differently. No single setting works the same way everywhere:

Claude Desktop: Set permissions per tool category, or per individual tool via the Custom setting, in the connector’s settings panel. See Claude’s documentation on connector permissions
ChatGPT: Asks for confirmation before certain actions by default. You can remember a choice for the rest of a conversation, but it resets when you start a new one. See OpenAI’s help center on apps in ChatGPT for more detail.
VS Code: Confirm you trust the server when first prompted, then approve tool calls as they run. Use the Configure Tools button in the Chat view to enable or disable individual tools, and MCP: Reset Trust from the Command Palette to re-confirm a server. See VS Code’s documentation on managing approvals and permissions for more detail.
Cursor: Choose a Run Mode under Settings > Agents > Approvals & Execution. Auto-review (the default) asks before risky calls. Allowlist and Run Everything skip approval entirely. See Cursor’s documentation on Run Modes for more detail.
GitHub Copilot CLI: Copilot prompts before running a tool, with options to allow it once or for the session. Run /mcp to review configured servers and their tools.

When in doubt, keep your client’s default approval behavior. It adds one confirmation step and helps you catch a mistake before it happens.

Tool	Type	Description
`mandatory_init`	Setup	Loads the Frame.io usage guidance and, by default, lists the first page of accounts you can access.
`read_context`	Read	Shows the signed-in user and helps you browse accounts, workspaces, and projects. You can choose what to view or open a specific account, workspace, or project by ID.
`search_assets`	Read	Searches an account for projects, folders, files, and version stacks by name or content.

Assets and structure

Tool	Type	Description
`read_assets`	Read	Shows a file, folder, or version stack, or lists the contents of a folder or the versions in a stack. File and version details can include media links.
`manage_assets`	Write	Moves or copies files, folders, and version stacks. Moving a file into a version stack adds it as a new version.
`manage_versions`	Write	Creates a version stack from 2–10 files in the same folder, or changes the order of versions in an existing stack.
`change_structure`	Write	Creates workspaces, projects, and folders; renames workspaces, folders, and files; or updates a project’s name, status, or access restriction.
`list_recently_deleted`	Read	Lists recently deleted files, folders, and version stacks in a project.
`restore_assets`	Write	Restores a recently deleted file, folder, or version stack to its original location.

Comments

Tool	Type	Description
`read_comments`	Read	Shows a file’s comments and replies, or the details of a specific comment.
`write_comments`	Write	Adds or updates a comment, marks it complete or reopens it, or attaches a file. Comments can be anchored to a video timecode, PDF page, or image position.

Metadata

Tool	Type	Description
`list_metadata_fields`	Read	Lists an account’s custom metadata fields, including their types and available options.
`change_metadata`	Write	Creates or renames a custom metadata field, or sets metadata values on files. A metadata update applies the same values to all selected files in one project.

Tool	Type	Description
`read_shares`	Read	Lists a project’s shares, or shows a share’s details, contents, or reviewers.
`manage_shares`	Write	Creates or updates a share, or adds or removes content from it. Public shares are accessible to anyone with the link.
`manage_reviewers`	Write	Lists, invites, or removes reviewers on a secure share. Inviting reviewers sends them an email.

Membership

Tool	Type	Description
`read_members`	Read	Lists people and their roles at the account, workspace, project, or folder level.
`manage_members`	Write	Changes a member’s role or removes their direct membership at the account, workspace, project, or folder level.

Uploads

Tool	Type	Description
`upload_media`	Write	Uploads a file from your machine or has Frame.io fetch a file from a public web link.
`get_upload_status`	Read	Checks a file’s upload progress and processing status.

Deletion and diagnostics

Tool	Type	Description
`delete_resources`	Delete	Deletes a file, folder, project, workspace, share, comment, or metadata field. Deleted files and folders can be restored; other listed resources are permanently deleted.
`read_audit_logs`	Read	Shows account activity, including who changed a resource and when. Results can be filtered by date, event, or resource.
`report_tool_issue`	Feedback	Sends feedback to the Frame.io MCP team about a problem or a difficult workflow with these tools.

Examples

Audit Logs + Reporting

Generate a list of everyone that’s logged into my Frame.io account in the last timeframe with the date and time they logged in.

Generate a report of all Frame.io share links created in the last timeframe, including the date and time created, which assets were included, who accessed the link, and whether it was a public or secure share.

What is the current storage utilization per workspace and per project in my Frame.io account?

Generate a report of all users that have uploaded content into my Frame.io account, including the name of each file, its file type, and which project or workspace it was uploaded into.

Generate a report of all approved files over the last timeframe, including who commented, who approved each file, and which project or workspace those files live in.

Upload my asset to Frame.io, generate a share link, and request a review from the following people.

Get all of the comments on my share.

Upload the new version of my asset and update the same share link.

Browse + Discover

Show me all the projects in my main workspace.

Find any files with ‘rough’, ‘draft’, or ‘wip’ in the name across my account.

Find the ‘Final Delivery’ folder in the ‘Fall Campaign’ project and list everything in it.

What’s the upload and processing status of everything in the ‘Final Deliverables’ folder of the ‘Q3 Campaign’ project?

Cross-account Overview

Which projects across my accounts had uploads or comments in the last two weeks?

What’s the status of every active project across all my accounts? Group by workspace.

Disconnecting

Each client has its own specific method for disconnecting or removing the Frame.io MCP:

Claude Desktop: Go to Customize > Connectors, find Frame.io, and remove it. See Claude’s documentation on connectors for more detail.
ChatGPT: There are two options. To disconnect (keep the plugin installed but stop its access), open Settings > Plugins > Frame.io and toggle it off; toggle it back on to reconnect. To remove it entirely, open Plugins in the sidebar, select Frame.io under Installed, and choose Uninstall. You will need to reinstall it from the plugin directory to use it again. See OpenAI’s help center on apps in ChatGPT for more detail.
VS Code: Remove the Frame.io entry from your MCP configuration. See VS Code’s documentation on managing MCP servers for more detail.
Cursor: Remove the frameio-mcp entry from ~/.cursor/mcp.json, the same file used to add it.
GitHub Copilot CLI: Run /mcp delete [server-name] to remove the server.

Troubleshooting

Configured the Frame.io MCP before October 7, 2026

Check that its server URL matches the current V1 URL: https://mcp.frame.io/v1/mcp

Update it if needed, then restart or reconnect the MCP server.

OAuth loop or authentication failure

Sign in with the Adobe ID linked to your Frame.io account. If you have multiple Adobe accounts, confirm you are using the correct one. Removing and re-adding the Frame.io MCP in your client starts a fresh OAuth session.

AMA multi-account authentication errors

If Adobe IMS silently reuses an existing browser session, you may authenticate into the wrong organization without being prompted to choose. The fix is clearing the Adobe session that’s being reused. Clear your browser’s cookies for Adobe’s login domain before reconnecting, or open a separate Adobe tab in the same browser and sign out completely from there. Once the previous session is cleared, you will be presented with a profile selection option at authentication, letting you choose the correct organization.

Frame.io MCP not visible or not connecting after adding

Verify the URL is exactly https://mcp.frame.io/v1/mcp. Restarting your client after adding it may be required.

Tools not available in conversation

Verify the Frame.io MCP shows as connected in your client. If it is connected but tools are still missing, check that the tool category is not set to Blocked under permissions.

Additionally,

Connected successfully but no workspaces or projects appear

You are authenticated but may have no workspaces in the connected account, or may be in the wrong account entirely. Ask your AI agent to list all accounts first. If you have access to multiple, you may need to specify a different account ID. AMA users should also verify they authenticated under the correct Adobe organization (see above).

Permission errors on specific tools

Tool calls reflect your Frame.io account permissions. Uploads need Contributor access. See Managing user permissions for details.

Upload stalled or asset stuck in processing

Ask your AI agent to check the status directly: “What’s the upload status of my recent file?” It will call get_upload_status. Large or high-resolution files can take several minutes to transcode.