Practical MCP workflows in DaVinci Resolve

“Read” describes the requested task, not enforced permissions. MCP tool annotations are hints; review the server, arguments and proposed code instead of trusting a label.[8]

1. One known environment

   Record Resolve version and edition, operating system, client and server versions. State whether the server is native or third-party. Do not assume this article’s protocol revision is what your installation uses.

2. One actual catalogue

   Use the chosen server’s tool definitions and account for all catalogue pages. Check whether the task is available through a bounded tool or requires a script that must be inspected before approval.

3. One precise target

   Specify the already open project, selected timeline and, where relevant, a bin, tracks or range. Establish units, origin and limits. Resolve ambiguity before calling anything.

4. An observable before-state

   Write down the values you expect to compare. Before the marker exercise, prepare a recoverable copy and a recovery method you have actually checked. Do not test recovery on the working original.

5. A stopping rule and an independent check

   State what must remain untouched and how the outcome will be checked in Resolve. After a write with an unknown outcome, inspect state before considering any new action.

Catalogue pagination and protocol revisions are specified by MCP; the five-step preparation above is our editorial checklist.[9][11]

1. Identify first

   Match the project and timeline to the intended targets. Read stable identifiers where available. Duplicate names are not permission to pick the first object.

2. Read available settings

   Request resolution, frame rate and start timecode for that timeline only where reading is supported. Record project settings separately from timeline settings.

3. Keep the gaps visible

   Mark an unavailable field as “not obtained”. Do not infer it from the project name, a random clip’s properties or an earlier conversation.

Limits. This passport establishes reading scope, not overall project health. It does not prove media availability, picture quality or readiness for delivery.

1. Bound the scope

   Name the exact bin and request direct children only. If reading would require changing the current selection or bin, agree on that separately first.

2. Minimize the fields

   Request names, available item types, durations with units and frame rates with sources. Do not force stills, audio or nested timelines into a video-clip model.

3. Report coverage

   State the number of returned items and the traversal limit. A limit or inaccessible page makes this a partial report, not an inventory of the entire Media Pool.

Limits. Matching names do not establish matching files. Metadata does not prove file existence, readability or byte-for-byte identity. File-system inspection and relinking are outside scope.

1. Establish coordinates

   Request original position values and their units. Do not assume 24 fps or round 23.976 or 29.97 to an integer. An unknown drop-frame setting is a reason to defer conversion.

2. Read only the agreed scope

   For each permitted track, obtain its available type, number, name and items with start and end values. Mark a nested composition as nested when its contents cannot be inspected.

3. Separate structure from interpretation

   An apparent gap on one track is not automatically a black frame or silence across the timeline. A file name alone is not evidence of its content.

Limits. This is a structural report, not a judgment of pacing, meaning, audio levels, grading or edit quality. Those require a separate viewing or analysis of actual media.

1. Compare against a brief

   For each requirement, record expected and actually read values. Label the source separately: timeline settings, output settings or a particular render job.

2. Do not fill in missing data

   Do not infer codec, colour conversion or audio configuration from a preset name. Leave inaccessible fields for manual inspection in Deliver.

3. Keep reporting separate from execution

   Do not apply a preset, save new settings, add jobs or start the queue. Destination, free space, write permissions and existing files need separate checks.

Limits. Matching settings do not prove encoding availability or the quality of a future file. This exercise does not start a render, change the queue or access disk to check space.

1. Plan first, without writing

   Read the target timeline and existing markers. Show the exact tool, arguments or complete script. Do not write until a person separately approves that exact plan in their client.

2. One approved write

   Immediately before acting, recheck the target and absence of conflict. Permit one new marker only. No replacing existing markers, clearing markers, editing clips, rendering or automatic retries.

3. Read the result again

   Read the markers after the operation. Compare before and after: exactly the requested object was added and existing markers stayed unchanged. Then check position and properties in the interface.

Limits. A plan or downloaded TXT is not client authorization. Do not repeat an operation after an unknown outcome without checking state. Restoring a copy or deleting only the new marker is a separately approved action; do not assume Undo reverses everything.