Your AI. Your engineer. Your IT, connected.
MCP gives your assistant access to the evidence and remote tools an engineer needs. Use ChatGPT or a compatible Claude client to investigate a problem, take an authorised action and verify the result in Vycko.
No Vycko AI tokens for direct MCP tools.
Keep the reasoning in your own AI account.
Your assistant reads the permitted device information, works through the problem and calls Vycko’s remote tools. These direct MCP requests use the platform’s API and device tools; they do not invoke Vycko’s built-in AI.
The conversation and model reasoning still use your ChatGPT or Claude account. Your provider’s subscription, usage limits, additional credits or API charges apply. MCP does not make model inference token-free or remove the Vycko service subscription. Vycko’s built-in AI is separate and has its own configured usage.
What is MCP, in practical terms?
The Model Context Protocol is a standard way for an AI assistant to work with authorised tools. Instead of copying a screenshot or log into a chat, an engineer can let the assistant request current device evidence from Vycko and use supported actions.
Vycko remains the connection to your managed computers. The assistant provides the conversation and reasoning. The engineer decides which customers and permissions the connection receives, and is responsible for reviewing the work.
Connect the assistant you already use.
ChatGPT: private OAuth connection
An administrator creates a private connection in Settings → API access, using the callback address supplied by ChatGPT. The administrator configures the client ID and secret in the assistant, signs in to Vycko, and approves the customer scope and permitted actions. The private connection can be turned off and its access revoked.
Use your own Vycko installation’s MCP address, ending in /mcp. The marketing website at vycko.com is not the MCP endpoint. ChatGPT’s plan, workspace permissions and current support for custom apps determine which read and write actions are available. Check OpenAI’s current MCP guidance.
Claude Code: scoped API key over HTTP
Vycko’s Streamable HTTP MCP endpoint accepts a scoped API key as a bearer credential. Claude Code supports HTTP MCP servers with an authorisation header. Create a narrowly scoped key, configure it securely in your Claude client, then confirm which tools are available. See Anthropic’s MCP connection documentation.
Claude Desktop and other clients
A compatible local MCP bridge can provide the bearer-authenticated connection for Claude Desktop. The current Vycko private OAuth callback allowlist supports ChatGPT and local clients; it does not currently provide a direct claude.ai remote OAuth connection. Confirm the client and authentication route with us before rollout.
Never paste a live API key or client secret into a general chat, a shared prompt, a public repository or a support screenshot. Use the client’s secure connection configuration.
Real evidence. Practical remote actions.
| Engineer’s task | Examples of available MCP tools |
|---|---|
| Find the right customer and computer | list_customers, search, list_devices, get_device_snapshot |
| Understand a fault | remote_services, remote_processes, remote_event_log, remote_list_folder, remote_read_file |
| Review monitoring evidence | list_alerts, get_fleet_health, get_device_telemetry, get_device_check_readings |
| Make an authorised repair | request_runbook, remote_service_control, file changes and, when explicitly enabled, remote_run_command |
| Track the work and verify it | get_job, incident history, follow-up device reads, resolve_incident |
| Transfer permitted files | Chunked download and upload tools, with file permissions, size limits and integrity checks |
These are examples from the supported tool catalogue. Dashboard features and MCP tools are not identical: some administration, security and scheduling operations remain available only through a signed-in dashboard. Device state, operating system, agent version and customer settings affect availability.
From “it won’t print” to a checked repair.
- Find the computer. The engineer identifies the customer and affected device, then checks its current state.
- Read the evidence. The assistant inspects the print service and relevant event logs. A symptom is a starting point, not a diagnosis.
- Choose a repair. The engineer reviews the proposed change. A reviewed runbook follows the platform’s configured approval policy.
- Act with the granted authority. The permitted action runs through Vycko. Offline devices, unavailable tools or refused permissions need a different next step.
- Verify and record. Check the job outcome and service state, test the user’s issue where appropriate, and add the result to the incident.
This is an example of the workflow, not a guarantee that restarting a service fixes every printing problem.
Give the assistant the access the job needs.
Start with a read-only key or connection, restricted to the relevant customers. Enable change rights only when needed. Console access and estate-management rights are separate capabilities; they should be granted deliberately.
Reviewed runbooks follow their applicable policy and approval rules. Direct remote changes can take effect immediately once the connection has change rights. The current change tool set also includes approval decisions made in the key owner’s name. Do not treat MCP as a guarantee that every write will wait for a separate human prompt: configure the scope carefully and review your assistant’s action-confirmation settings.
Requests are subject to the platform’s access checks and audit trail. Keys and connections can expire or be revoked. A log supports accountability; it does not make an unsafe command safe or replace the engineer’s judgement.
Understand what your assistant can see.
Data returned by a tool becomes available to the connected assistant and may include device names, usernames, paths, incident notes, logs and requested file content. Your chosen AI provider handles that information under its own terms, privacy settings and account controls.
Confirm the customer’s authority and your organisation’s AI data-sharing rules before connecting. Limit the scope, avoid unnecessary personal data, and review retention and model-training settings with the provider. See our privacy notice and acceptable-use guidance.
See the workflow in your own environment.
Tell us which assistant you use, which computers you support and what you want to diagnose. We can demonstrate the available tools, check the connection route and explain the permissions before you enable device changes.
Call 01636 362361 or email info@vycko.com. ChatGPT and Claude are third-party products; naming them describes compatibility and does not imply endorsement or partnership.
See Vycko in your environment.
Tell us about the computers or clients you support.
Newark Solutions Ltd trading as Vycko