What is MCP? Servers, clients, tools and resources in plain words

Model Context Protocol (MCP) is an open protocol that lets AI applications connect to external tools and data through a common interface. An MCP server is a program that exposes those capabilities; an MCP client is the component inside an AI application that communicates with it. The official specification defines the interface, while the connected software determines what you can actually access or do.

What it can do

MCP can give an assistant access to information or actions beyond the text you paste into chat. Its usefulness depends on the server, your account permissions and the client’s supported features. The following are illustrative workflows, not claims that every server includes these operations:

• For a support question, retrieve an approved troubleshooting document and ask the assistant to identify the relevant passage.
• For a database question, provide the table schema as context before requesting an explanation of a query.
• For a coding task, retrieve issue details before drafting a change, with permission to publish anything handled separately.

Start with a task whose result you can inspect. Asking for a document title and a relevant paragraph gives you something to compare with the original. A broad request to manage a project makes it harder to see whether the right information was retrieved or whether an action exceeded your intent.

For specific integration targets, see the GitHub MCP guide and Postgres MCP guide. More beginner explanations belong to AI agents explained.

Hosts, clients and servers in plain words

The host is the AI application you use. The client is its connector to an MCP server. The server supplies the capabilities. A server can run on your computer or elsewhere: server describes its role, not a requirement to rent a remote machine. VS Code is an example of a host in the official architecture overview.

Think of a document request moving through the chain: you ask the host, its client communicates with the configured server, and the server returns information for the host to use. When diagnosing a problem, identify where that chain stops. A working chat window alone tells you little about the server connection.

Tools, resources and prompts

These server features have different jobs:

A tool can retrieve information as well as change something. Resources provide read-only context, while prompts help structure an interaction. The official server guide explains these distinctions; individual servers decide which features they expose.

For your first evaluation, write down the input, expected output and any possible side effect. For a document lookup, an acceptable result might include the document identity and an excerpt you can verify. For a change request, record the intended destination and require a separate review before submission. This makes tool selection a practical decision rather than a name-recognition exercise.

Current specification and transport types

As of 30 Oct 2026, the official versioning page identifies 2026-07-28 as the current protocol version. The date identifies a specification revision; it is separate from an extension or server package version. Older implementations can support earlier revisions, so check compatibility for the actual client and server you choose.

The current transport specification defines:

• stdio: messages pass through the standard input and output streams of a subprocess launched by the client. This is the local-process option.
• Streamable HTTP: messages go to an MCP endpoint using HTTP POST; responses can be JSON or an SSE stream. This supports communication with a separately running service.

Custom transports are also possible. Check the current documentation before treating a legacy SSE setup guide as interchangeable with Streamable HTTP. Match the publisher’s endpoint or launch command to the transport your client supports.

How to connect it safely

For VS Code, follow the official connection walkthrough. Other editor clients can use different settings and approval interfaces.

1. Choose a narrow task. Use a non-sensitive document lookup for the initial evaluation. Decide what evidence would show that it worked.
2. Review the publisher and requirements. Inspect the endpoint or launch command and requested access. A local command runs software on the machine; a remote service may require sign-in.
3. Add the server. In VS Code’s Command Palette, run MCP: Add Server. Choose the HTTP option for a published endpoint or Command (stdio) for a launch command. Use the publisher’s documented values.
4. Choose the configuration destination deliberately. Follow the options offered by your selected agent harness. Keep credentials out of chat messages and committed project configuration.
5. Review trust and tool access. Enable only the capabilities needed for the task. Read authentication and approval prompts before accepting them.
6. Inspect an actual call. Ask explicitly for the intended server and a low-impact operation. Expand the tool output and compare it with the final answer. A plausible response alone does not demonstrate a working connection.

Before adding sensitive data or write operations, use the MCP server security checklist. Keep a small record of what you enabled so a later configuration change is easier to review.

When this does not work

The connection fails. A missing runtime, incorrect command, wrong endpoint or incomplete sign-in can stop the workflow before any useful result. Follow the client’s startup diagnostics and the server publisher’s instructions rather than repeatedly submitting the same prompt. VS Code’s connection guide includes troubleshooting paths.

The capability is missing. The host, server and selected account must support the operation you need. MCP does not dictate how an application uses the supplied context. Check the exposed features before assuming that a server name means access to an entire service. See MCP versus APIs and function calling for the interface distinction.

Trust is misplaced. A standardized interface does not enforce a sandbox or make tool descriptions trustworthy. The specification’s security principles require attention to consent, privacy and tool execution. Review returned material as evidence; do not treat instructions inside a retrieved document as authorization to expand the task.

Where Tvijo fits

As of 30 Oct 2026, the Tvijo MCP release status describes a published consultation server and prepared supplied-diff review functionality awaiting publication. Built-in editor integration and delegation remain planned, with separate acceptance gates.

The DSHREM VS Code extension is an editor-side Tvijo client. Tvijo MCP is the server-side example that a compatible MCP client can call. These roles illustrate the client/server distinction, but the extension’s existing Tvijo connection does not establish that its planned MCP integration has shipped. For an evaluation, check the package available to your client rather than assuming prepared source code is already distributed.

TvijoCode packages the extension in a Code - OSS editor. Its documented status is a Windows x64 test build, unsigned and without automatic updates. Installing that editor does not complete an MCP connection.

As of 30 Oct 2026, the Tvijo workspace lists Try at $0: sponsored access ends at whichever comes first—5 tasks, 50,000 tokens or $0.05 of usage—and draws on a shared $1 daily pool funded by TVIJO. Service credit is shown as $10 for requesting prepaid credit, with metered provider usage plus an 8% service markup. Payment collection is not activated; an order is not a charge.

FAQ

Does an MCP server need a separate machine?

No. It can run locally or remotely. Check whether its documented connection uses a local launch command or an HTTP endpoint.

Is every tool a write action?

No. Tools can retrieve data too. Inspect the operation and permissions instead of deciding from the word tool alone.

Does adding a server prove it works?

No. Verify a call to the intended server, inspect the returned output and check that the answer reflects it.

Model Context Protocol is a trademark of its owner. Tvijo is not affiliated with or endorsed by it.