Workbenches

Workbenches: your own code, running inside LabMission

When your lab needs something no application has, write it, or have an agent write it, and run it as a workbench. It deploys into your workspace, answers to the same access rights, and is available to people, applications and AI agents.

Workbench · label-serviceRunning · 2 replicas
Source
github.com/your-lab/label-service @ v1.4.0
Region
West Europe
Resources
0.5 CPU · 1 GB · 2 replicas
Public traffic
Closed
Consumes
printer-registry
Secrets
From your key vault
Modes
API
Requires: Labels.Print
MCP
Tools: print_label, void_label
How a workbench works

From a repository to a governed service in your workspace

Point a workbench at your code. LabMission runs it in your cloud with the resources you set, and exposes it through the modes you choose, each with its own sign-in and required rights.

YOUR CODEWORKBENCHMODESUSED BYGit repositoryBranch or tagPackageVersioned releaseWritten byA person, an agent, or bothWorkbenchRuns in a container in your cloudYou set region, CPU, memory, replicasSecrets come from your key vaultCan consume other workbenchesCan be closed to public trafficAPI modeOwn sign-in andrequired rightsMCP modeAgents call it as tools,under the same rightsPeople & appsAI agentsOther workbenchesDeployed through the same proposal, review and deploy path as every other change, and recorded.
Five steps

How a workbench reaches production

  1. 1

    Point at the code

    A Git repository at a branch or tag, or a versioned package.

  2. 2

    Set the runtime

    Region, CPU, memory and replicas, and whether it is closed to public traffic.

  3. 3

    Choose its modes

    Expose it as an API, as MCP tools for agents, or both, with the rights each requires.

  4. 4

    Propose and approve

    The workbench arrives as a proposal and is reviewed like any other change.

  5. 5

    Use it

    People, applications, agents and other workbenches call it, under their rights.

What labs build

Workbenches a lab might run

Each of these is a small, specific piece of code. The workbench gives it a home, an identity and a record.

Label printing service

Formats and sends labels to the printers on each bench, from any application.

Consent form intake

Reads signed consent forms and attaches the reference data to the samples they cover.

Instrument bridge

Speaks an instrument's own protocol and forwards readings into a connection.

Calculations and QC rules

Your lab's own calculations, run the same way every time and recorded.

A toolset for your agents

Domain tools, exposed in MCP mode, that your agents call alongside the platform's own.

Analytics hand-off

Prepares extracts for a data platform or a statistician, on a schedule or on request.

A gloved scientist applying samples with a micropipette.
Workbenches and agents

An agent can write the workbench, and then use it

Ask an agent for a tool your lab needs. It writes the code, proposes the workbench, and once you approve it, calls it in MCP mode like any other tool. A person still decides what runs.

  • The code lives in your repository, where your team can read it
  • The proposal shows what will run and with which rights
  • MCP mode makes the result a tool for every agent you allow

Bring a tool your lab needs

Tell us about the gap. We'll show how it becomes a workbench in your workspace, written by your team, an agent, or both.