AI & Machine Learning · VLM Run + Claude
VLM Run MCP with Claude: Structured Data From Documents and Images
The short answer
Claude connects to VLM Run through the Model Context Protocol, an open standard that lets Claude use software securely. Once connected, Claude can ask VLM Run to extract structured data from documents and images, run multimodal predictions, manage files and review evaluation results, all through natural language instructions with permissions your team controls.
What is VLM Run?
VLM Run is a multimodal AI platform for teams that work with documents, images and other visual information. It offers tools for structured extraction, predictions, file handling, skills, feedback and evaluations. Teams use it to build vision-language workflows that turn visual material into reliable, usable data, without assembling model, file and evaluation infrastructure by hand. It suits operations, data and product teams building document-heavy processes.
How Claude connects to VLM Run through MCP
The Model Context Protocol, or MCP, is an open standard that lets Claude use other software securely. Think of it as a universal adapter between Claude and a tool. VLM Run offers its own MCP server, and it can also be reached through its direct API, so your team can choose the route that suits its systems.
Once connected, Claude can work with VLM Run in plain language. A staff member might ask Claude to pull named fields from a batch of scanned forms, run a prediction on a set of images, organise files or check how recent outputs scored in an evaluation. Claude carries out these requests using the access you grant, typically through an API key.
Permissions and approvals keep people in charge. Access is limited to what each workflow needs, and sensitive actions such as deleting files or publishing results can require a human sign-off. Activity can be logged so that every request and response can be reviewed later.
What you can automate
Invoice and receipt data capture
Finance teams often retype figures from supplier invoices into accounting systems. With VLM Run connected to Claude, invoices can be read for supplier name, dates, line items and totals, returned as structured data, and checked against purchase orders before anyone posts them. Staff review only the exceptions, which typically reduces manual keying and speeds up month-end close.
Identity and onboarding document checks
HR and compliance teams review passports, licences and certificates during onboarding. Claude can ask VLM Run to extract the relevant fields from each upload and flag missing or unreadable documents for follow-up. A person still makes the final decision, and the extracted data can be stored with an audit trail to support internal checks.
Quality inspection from images
Manufacturing and logistics teams photograph goods, labels or damage at key stages. Claude can request predictions from VLM Run on those images, classify the condition, and summarise recurring issues for the quality lead. This helps teams spot patterns earlier, while inspectors keep authority over release decisions.
Evaluating document AI accuracy
Product and data teams need to know whether extraction quality is improving. Claude can retrieve evaluation results from VLM Run, compare recent runs against earlier ones and write a short summary of where errors cluster. This gives non-specialists a clear view of model performance without learning the platform's interface.
Claims and case file processing
Insurance and customer service teams receive mixed bundles of forms, photos and letters. Claude can organise these files, extract the key facts from each, and prepare a structured case summary for a handler. The handler confirms the details and decides the outcome, which keeps judgement with the people accountable for it.
An end-to-end workflow: Supplier invoice intake to accounts payable
TriggerA supplier invoice arrives in a shared finance mailbox or document folder.
Claude detects the new invoice and passes the file to VLM Run for structured extraction.
VLM Run returns supplier, invoice number, dates, line items and totals as structured data.
Claude compares the extracted figures with the matching purchase order held in the finance system.
Matching invoices are queued for payment; mismatches are flagged with the specific differences noted.
A finance clerk reviews flagged items and approves or corrects them before posting.
The processed file and its extracted data are stored with a log of each step.
OutcomeTypical results include less manual data entry, faster processing of routine invoices, and a clear record of every exception for audit. Exact gains depend on invoice volume and existing processes.
How we set it up
Free automation audit
We review how your team handles documents, images and evaluations today, and identify the workflows worth automating with VLM Run and Claude.
Secure connection
We connect Claude to VLM Run using the API key sign-in method, storing the key in a secure credential store and granting least-privilege access to only the functions each workflow needs.
Workflow design
We design the end-to-end process across VLM Run and your other systems, such as Microsoft 365, CRM, ERP or ticketing, and agree which actions require human approval.
Testing with real cases
We test the workflow against past documents and images from your own business, measure extraction accuracy and tune prompts and checks before anything goes live.
Launch and training
We launch with monitoring and audit logs in place, then train your team on how to use, review and escalate the automated workflow.
Who benefits
- Finance director
- Fewer manual keying errors and a faster, better-documented invoice process.
- Operations manager
- Routine document and image handling runs on a schedule, freeing staff for exceptions.
- Head of compliance
- Structured extraction logs and human sign-offs give a clear audit trail.
- Product or data lead
- Evaluation results become easy to query and summarise in plain language.
Security and control
- Least-privilege API access, limited to the functions each workflow needs.
- Human approval required for consequential actions such as deleting files or posting records.
- Audit logs of requests, responses and approvals for review.
- Testing against your own historical cases before launch, with ongoing accuracy checks.
- Clear escalation paths when extraction confidence is low or documents are unreadable.
For teams in the US, UK, India and the Middle East
Organisations in the United States and the United Kingdom can host the workflow in their chosen region to meet local data requirements. In India, we design workflows to meet the Digital Personal Data Protection Act, 2023. For organisations in the UAE and Saudi Arabia, we align data handling with local data protection laws. Hosting region is chosen during design so that it meets your specific legal and contractual requirements.
Learn more about Claude and MCP
Anthropic's AI model and assistant.
Secure links between Claude and business tools.
How engineers build and change software with Claude.
Hosted agents that run multi-step work.
The open standard Claude uses to connect to tools.
Frequently asked questions
What does the VLM Run MCP server do?
The VLM Run MCP server lets Claude use VLM Run's capabilities through a standard, secure connection. Claude can request structured extraction from documents and images, run multimodal predictions, manage files and review evaluation results. It works alongside the direct API, so your team can choose the route that fits its systems.
Is connecting VLM Run to Claude secure?
Security depends on how the connection is set up. We store the API key in a secure credential store, grant only the access each workflow needs, log activity and require human approval for sensitive actions. We also review data handling and hosting region against your legal obligations before launch.
What can Claude do with VLM Run once connected?
Typically, Claude can ask VLM Run to extract structured fields from documents, run predictions on images, organise and retrieve files and summarise evaluation results. Users make these requests in plain language. The exact actions available depend on the access you grant and the workflow we design.
How long does setup take?
Timelines vary with the number of workflows and the complexity of your systems. A focused single workflow often moves from audit to launch in a few weeks, including testing against your own cases and training. We confirm a realistic timeline during the free automation audit.
Which plan or permissions do I need?
You need an API key from your VLM Run account with permissions covering the functions your workflow uses. The specific plan and access level depend on your usage and on your VLM Run account terms, so we check these during the audit and recommend the least-privilege option that meets your needs.