Claude Artifacts and MCP: Connect Tools to Your Work

Claude Artifacts and MCP (Model Context Protocol) serve different purposes, and they do not overlap in the way you might expect. Artifacts are a rendering feature inside Claude.ai: Claude produces a self-contained HTML, code, or document output that you can preview and share. MCP is a connection protocol: it lets Claude call external tools, read files, and pull in live data during a session. Tools like fenbs use MCP to give Claude shared project context — decisions, tasks, and where the work stands — that travels across sessions and tools.

This article explains what Artifacts can and cannot do, when MCP is the right choice instead, and how to decide which Claude feature applies to what you are trying to accomplish.

What Claude Artifacts actually are

An Artifact is a rendered output that Claude creates alongside a conversation. When Claude writes a React component, an HTML page, or a markdown document, it can present it as an Artifact: a side panel that shows the rendered output rather than raw text. You can share an Artifact link and the recipient sees the rendered version without a Claude account.

Artifacts are stateless and sandboxed. They do not call external APIs on their own, they do not have access to your files, and they do not persist between conversations unless Claude regenerates them. They are a presentation layer, not an integration layer. If you share an Artifact, you are sharing a snapshot of what Claude produced, not a live connection.

What MCP adds that Artifacts do not provide

MCP is a protocol that lets Claude clients connect to servers that expose tools, resources, and prompts. A Claude client with an MCP connection can call tools during a conversation: read a file from your system, query a database, get the current state of a project board. The results come back into the conversation and Claude uses them in its answer.

This is fundamentally different from Artifacts. Artifacts render output that Claude already produced. MCP lets Claude pull in live external data before producing the output. The distinction matters when you are deciding how to connect Claude to your tools and data.

Which Claude clients support MCP

MCP requires a client that implements the protocol. Claude Desktop supports MCP servers configured locally. Claude Code and the API support MCP with appropriate setup. Claude.ai in a browser, where Artifacts appear, does not run local MCP servers. If you need MCP tool calls, check whether the Claude client you are using supports the protocol.

• Claude Desktop: supports MCP servers configured in its settings; these run alongside the desktop app.
• Claude Code: supports MCP tools in the editor context.
• Claude API: MCP can be integrated into applications built on the API.
• Claude.ai browser: check current documentation; MCP support for the web interface changes as Anthropic updates the product.

Claude Projects and MCP are separate things

Claude Projects (available on Claude.ai) let you set a persistent system prompt and upload files that are available across conversations. This gives persistent context within the Claude.ai interface. MCP gives tool-calling ability and live data access in clients that support the protocol. They are complementary: a Project system prompt can instruct Claude how to use MCP tools, but Projects alone do not give Claude the ability to call those tools.

fenbs connects over MCP. When a Claude client has fenbs configured as an MCP server, Claude can read the project's decisions, tasks, and context notes at the start of a session. This is separate from Projects; it works in any MCP-capable client, not just Claude.ai.

How to do this with fenbs

Use Artifacts for shareable rendered outputs and fenbs over MCP for shared project context that any connected assistant reads at the start of a session.

1. Create a free account and add a project for the work you are managing with Claude.
2. Connect a supported MCP-capable Claude client to fenbs via Integrations to give Claude access to project context during sessions.
3. Add the decisions, tasks, and current state notes for your project in fenbs so Claude reads them at session start rather than requiring re-explanation.
4. Use Artifacts for outputs you want to share or preview: rendered code, documents, or UI components.
5. When a session produces a confirmed decision or a new task, record it in fenbs so the next session — on any connected tool — starts from the current state.

When to use each feature

Use Artifacts when you want to share a rendered output — a working UI prototype, a formatted document, a visualization — with someone who does not need to see the conversation that produced it. Use MCP when you need Claude to read live data, call tools, or access external context during a conversation. Use Projects when you want a persistent system prompt and uploaded reference files available across Claude.ai sessions.

These features do not replace each other. A single Claude session might use an MCP tool call to read the current project state from fenbs, produce an Artifact with a formatted report, and save a new decision back through MCP. Each feature does its specific job.