MCP server reference
The Vidmatic MCP server is hosted at https://api.vidmatic.ai/mcp. This page is the reference for its MCP server tools, access levels and keys. Setup for each client is on the MCP server page; if something does not work, see Troubleshooting or the quickstart.
Connect
claude mcp add --transport http --scope user vidmatic https://api.vidmatic.ai/mcpThen /mcp → vidmatic → Authenticate. Cursor setup: /mcp#cursor. Cursor's web and background agents are not supported.
Access levels
When you sign in, the consent page asks for one access level.
| Access level | What your agent can do |
|---|---|
| Record & publish demos (default) | Read your recordings, upload recordings, add narration, publish demos you ask for, and report on Agent Workstreams |
| Read only | Read your recordings |
Browser sign-in and API keys
A browser sign-in shows up in Settings → Developer under Connected apps. Disconnect ends it at once. It renews itself while you use it, and lapses after 30 days without use.
If you prefer a key, press Build my command in Settings → Developer. You get the same one-line command with your key in it. Restart Claude Code once after you run it. Keys are listed separately: you can have up to 20 active keys, each lasting at most 365 days.
MCP server tools
| Tool | What it does |
|---|---|
get_skill | Returns a skill's instructions as Markdown |
get_recording_intel | Reads a recording: transcript, marks and frames |
get_annotation_image | Returns the image of one mark |
begin_upload | Starts an upload for a local video file |
preflight_narration | Checks that each spoken line fits its time slot, without writing anything |
create_demo | Turns an uploaded video into a narrated demo |
get_demo_status | Polls a demo until its narration is ready |
submit_narration | Replaces the narration script on a recording |
list_voices | Lists the narration voices |
start_workstream, report_phase, get_workstream, list_workstreams, get_stage_records | Drive and read Agent Workstreams; get_workstream can read one issue's card with issue= |
kickoff_from_recording | Links a recording to the work it starts |
report_requirement_status, post_workflow_progress | Report progress on one requirement, or post a live progress line |
get_optimization_catalog, report_optimization | Used by the optimize skill |
dispatch_agent, get_assignment, report_eval, complete_agent, fail_agent, rearm_agent | Used by the auto-implement-beta skill (given no issue URL, it works every open issue in the repo) |
report_events | Sends several workstream reports (stage changes, eval results, progress lines) in one call |
next_change_request, get_change_request, renew_change_request, resolve_change_request, list_change_requests | Used by the dispatcher skill to claim a queued change request, read it, and file it in the right repo |
report_dispatcher_note | Used by the dispatcher skill to tell you in your Inbox when a repo has work it cannot take on |
enqueue_change_request | Sends one of your recordings to your change request queue |
enqueue_change_request_text | Sends a change request written as text to your queue |
whoami | The account (email, plan, key) behind this connection |
Skills
The server also serves skills: step-by-step instructions your agent follows. Nothing is copied into your repository. See Skills.
A tool is missing
The server does not push tool-list changes to a running session. Reconnect vidmatic from /mcp, or restart your client.