Trusted Tech Leader/ Case studies

Case study · AI & automation

MCP server for AI-assisted video content workflow

An MCP server that lets an AI assistant run a multi-channel video pipeline, from research to script, visuals and voiceover.

Sector
Video content production
Category
AI & automation
Maturity
Working local system with an MCP server, a command-line companion and agent modules

At a glance

Who it's for
Video creators and small content teams running one or more channels.
Core stack
Python, Model Context Protocol, Google Sheets, Docs and Drive
What sets it apart
The assistant supplies the thinking and the MCP layer supplies structured access to the team's spreadsheet and documents.
On this page

What it looks like in use

Ask the assistant in plain words and watch the research, script and narration land in your own files.

Chat with the assistant: Channel "Sample Channel"
Conversation
Research a new idea about property tax appeals, then write the script for it.
Checked the sheet for similar topicsNo duplicate
Gathered web, news and forum materialDone
Saved the research brief to the cloud docsSaved
Wrote the script in this channel's toneSaved
Now make the narration audio.
Narration made on this computer and filed in the channel folderLinked in sheet
What needs a script next?
Standard deduction basics is waiting at Research CompleteFound
Result: the channel's pipeline sheet
Property Tax AppealsScript Complete
Standard deduction basicsResearch Complete
Filing late: what happensResearching
Script saved in Docs
Property Tax Appeals: a simple walk-through
Illustrative screen sample data, not client data

From a topic idea to a finished script and voice

  1. You ask in plain words

    Type what you want, such as researching a new topic for one of your channels.

  2. The assistant checks for repeats

    It looks through your sheet first, so you do not make the same video twice.

  3. It gathers the research

    Search results, news and forum talk are pulled into one short brief.

  4. Script and voice are made

    The script follows your channel's style, and a narration file is generated on your computer.

  5. Your sheet stays up to date

    Each topic's status and links are saved in the spreadsheet your team already uses.

The challenge

A video channel needs a steady pipeline of topics that move from idea to research, script, visuals, voiceover and publishing.

Teams usually track this in spreadsheets while the writing happens elsewhere, so status drifts, duplicate topics creep in and every channel reinvents its own prompts. Running several channels multiplies the problem: each has its own tracking sheet, storage folder, tone and prompt rules.

The solution

Responsibility is split cleanly. The AI assistant does the reasoning and writing, while an MCP server exposes tools that read and write the data layer: a Google Sheet as the pipeline database, Docs for research and scripts, and Drive for assets.

A web research layer gathers search, news and forum material into a brief. Each channel has its own sheet, folder and prompt set, and every tool accepts a channel parameter. A separate local text-to-speech tool runs as a subprocess to produce narration, which is uploaded to Drive and linked in the sheet.

For your technical teamhow it works under the hood

One topic from request to narration

Sequence of tool calls for one topicSequence diagram with six participants: creator, AI assistant, MCP server, research sources, Sheets Docs and Drive, and a local text-to-speech tool. The creator asks for a topic. The assistant has the server add it if it is not a duplicate, which writes a row to the sheet. The assistant gathers research through the server, which queries the sources and returns raw results, then saves a research doc that goes to Drive and updates the row. The assistant loads script context, saves a script doc, and the server stores it and updates the row. For narration the server runs the local text-to-speech tool as a subprocess, receives an audio file, uploads it and links it in the row.CreatorAI assistantMCP serverResearch sourcesSheets, Docs,DriveLocal TTS tool1. Asks for a topic2. add topic if new3. add row to the sheet4. gather research5. search sources6. raw results7. save research doc8. doc to Drive, row updated9. script context10. save script doc11. script to Drive, row updated12. make voiceover13. run TTS subprocess14. audio file15. upload audio, link in row
Time runs downward. The assistant does the thinking and writing; the MCP server does the reading and writing of data. Dashed arrows are replies. Narration is the only step that leaves the server for a separate local tool.

Key features

What the system does, as built in the codebase.

  • 01

    Multi-channel registry

    Add, switch, update and remove channels, each with its own sheet and Drive folder.

  • 02

    Pipeline tracking

    List, add and update topics, with workflow summary, stats and schedule views.

  • 03

    Duplicate detection

    Fuzzy matching runs before a new topic is added.

  • 04

    Research brief builder

    Web search, page fetch, forum and news search are combined into one brief.

  • 05

    Per-channel prompt sets

    Research, script, visuals, thumbnail, title and description prompts are loaded per channel.

  • 06

    Script, visuals and voiceover stages

    Context builders and save tools for each stage, with narration audio uploaded to Drive.

  • 07

    Performance insights

    Syncs platform statistics and surfaces category performance and series candidates.

  • 08

    Command-line companion

    Auth, setup and content commands that work outside the assistant.

How it fits together

From a creator's request to a script, visuals and narration.

  1. AI assistantreasoning and writing
  2. MCP serverroughly forty tools
  3. Agentsresearch, script, visuals, voice, analytics
  4. Research sourcessearch, news, forums
  5. Google Sheetspipeline state per channel
  6. Docs and Driveresearch, scripts, audio
  7. Local TTS toolnarration subprocess
The assistant calls MCP tools. Agents use services to gather research and to read and write Sheets, Docs and Drive. The voice agent shells out to a local text-to-speech tool.
  • Assistant calls MCP tools; agents use services to read and write Sheets, Docs and Drive.
  • The voice agent shells out to the local text-to-speech tool.

Implementation notes

Engineering decisions worth knowing about.

  1. Broad tool surface

    Roughly forty tools over a stdio server let any MCP-capable assistant drive the pipeline.

  2. Layered structure

    Agents orchestrate, and services wrap each Google API and research source.

  3. Sheet as state machine

    Sheet row status doubles as the workflow state, with colour-coding applied by a tool.

  4. Isolated text-to-speech

    TTS runs in a separate virtual environment through a subprocess, with a timeout, because of dependency incompatibility.

  5. Cached research

    Research results are cached with a TTL, and Brave search falls back to a keyless search.

  6. Per-channel auth and logging

    OAuth uses per-channel token files, with structured logging via structlog.

Technology stack

Taken from the project's dependency and config files.

Language
  • Python
Protocol
  • Model Context Protocol (mcp)
Google APIs
  • Sheets
  • Docs
  • Drive
  • YouTube data
CLI and logging
  • click
  • rich
  • structlog
Research
  • requests
  • httpx
  • BeautifulSoup
  • thefuzz
For your technical teamtechnology layers

Technology stack by layer

Technology stack layersLayered view: the Model Context Protocol on top, Python for the server and agents, research libraries for web sources, Google Sheets, Docs, Drive and YouTube data APIs for pipeline data and assets, and click, rich and structlog for the command line and logging.ProtocolTool interfaceModel Context Protocol (mcp)LanguageServer and agentsPythonResearchWeb sourcesrequestshttpxBeautifulSoupthefuzzGoogle APIsData and assetsSheetsDocsDriveYouTube dataCLI, loggingTerminal companionclickrichstructlog
The protocol layer is what the assistant sees. Python agents and services sit beneath it, reaching out to web research libraries and the Google APIs that hold the pipeline data.

Business value

What the system enables. No numbers are claimed.

  • Keeps AI-written content tied to a single source of truth for status.
  • Helps avoid duplicate topics and steers each channel's voice through its own prompts.
  • Lets one setup serve several channels without code changes.
  • Puts the stages from approved idea to script, visuals and narration into one assistant-driven flow.

Need a system like this built?

Tell us what it needs to do and we will scope it with you on a call.

Discuss a build