CI
Back to the work

MuseLabsMCP

I extended creative tools for agents, from AI provider integrations to blog-to-social workflows and OAuth refresh.

Platform & integration engineering
TypeScript · MCP · OAuth · Provider and job orchestration

  1. RequestUser & workspace

    Resolve access and the destination for the work.

  2. ExecutionTool & provider job

    Keep the job, progress, and returned files connected.

  3. ResultA file or draft

    Return a usable output with its status and cost.

Authentication, job records, and credit accounting support the full request.

Architecture illustration · Some tools wait for their provider; others can return a job to follow.

Building the integrations

An AI assistant can ask for an image, convert media, or turn an article into social posts. Making that request useful takes more than a callable function. The service has to know who is asking, follow work that takes time, account for its cost, and return something the person can use.

MuseLabs MCP connects creative capabilities to that full path. I contributed platform and integration work across AI tools, authentication, and content workflows. My work included updating the image and blog tools’ AI provider integration, extending blog-to-social production, and repairing OAuth refresh behavior within the shared platform.

Following the work

An authenticated request enters through the MCP server with a user identity attached. Media jobs retain a record of their state and outputs, while status tools and polling make ongoing work inspectable. The audio interface turns that state into a track list: which files are converting, which are ready, and where to download them.

That connection matters when a provider finishes after the original interaction has started. A useful result carries more than a success message. It includes the files, their titles and formats, the job status, and the credits charged. Storage-backed delivery gives the assistant and the person a practical next step.

I extended the same idea into M4 Studio’s blog-to-social workflow. The tool passes the source article and its published URL into a selected workspace, follows the production job, and returns draft posts for review. I also worked on workspace resolution and clearer errors, so account and workspace mismatches could lead to an actionable response.

Implemented audio interface · Rendered offline with demonstration tracks. No live conversion or charge is represented.

MuseLabs embedded audio interface with two example tracks ready to download and a third still converting

MuseLabs embedded audio interface with two example tracks ready to download and a third still converting

Implemented audio interface · Rendered offline with demonstration tracks. No live conversion or charge is represented.

Accounts, completion, and cost

The platform checks credit balances before starting paid work and records charges against jobs. For audio conversion, a stable provider-run reference links a completed job to its debit. Credit purchases use a checkout-session reference, allowing repeated payment notifications to resolve against the same ledger entry. These mechanisms keep cost attached to the work that produced it.

My OAuth work addressed another part of continuity: refreshing access as clients continue using their tools. Alongside job state, result delivery, and account controls, authentication helps the service remain usable across more than one request.

The outcome is a creative tool platform with a visible path from request to usable result. My contribution was in extending and connecting that path, including the less visible handoffs that determine whether an agent can finish the job it started.