Conversational Development With Claude Code — Part 14: Installing MCP Servers in Claude Code
TL;DR
MCP (Model Context Protocol) transforms Claude Code from a powerful reasoning engine into an operational collaborator. In this chapter, we install and integrate three production-grade MCP servers — Playwright, Notion, and Linear — and explore how extended context, explicit permissions, and transport protocols reshape professional development workflows.
This is not about plugins.
This is about expanding cognition.
From isolated reasoning to extended capability
Out of the box, Claude Code reasons over:
- Your conversation
- Your repository
- Your terminal instructions
That is powerful — but bounded.
Modern software engineering does not live inside a single directory.
It lives in:
- Product documentation (Notion)
- Task tracking systems (Linear)
- Browser-based validation environments (Playwright)
- External APIs
- Operational dashboards
MCP is the bridge.
It is Anthropic’s Model Context Protocol — a formal standard for extending a model’s capabilities through secure, permissioned servers.
When you install MCP servers, you are not adding features.
You are expanding the assistant’s operating surface.
What MCP actually is (and why it matters)
An MCP server exposes three primitives:
- Tools — executable actions (e.g., navigate to URL, create task, fetch document)
- Resources — structured external data (e.g., JSON documents, issue lists)
- Prompts — reusable structured interactions
These are not arbitrary extensions.
They are:
- Permission-gated
- Explicitly approved
- Session-bound
- Transport-defined
Claude Code becomes capable of acting beyond static reasoning — but only within clearly defined guardrails.
This is controlled agency.
Managing MCP from the command line
Claude Code exposes MCP management directly via CLI.
Basic commands:
cloud mcp help
cloud mcp list
help explains how to:
- Add a server
- Serve a local MCP
- Configure transports
list shows:
- Connected servers
- Status (connected / disconnected)
- Transport type
- Authentication state
This is infrastructure-level integration, not UI decoration.
Transport protocols: how MCP communicates
Claude Code supports three transports:
- HTTP
- Server-Sent Events (SSE)
- Standard Input / Output (stdio)
Each has different trade-offs:
- HTTP is common for hosted services.
- SSE is ideal for streaming event-driven systems.
- stdio is practical for local tool execution.
Understanding transport is part of professional configuration — especially in enterprise settings.
Installing the Playwright MCP Server
Playwright MCP gives Claude Code a browser.
Not metaphorically.
Literally.
It enables:
- Navigating to URLs
- Executing scroll
- Capturing screenshots
- Recording video
- Reading console errors
Installation is copied from the official Playwright MCP documentation under Claude Code integration.
Once added, verify:
cloud mcp list
You should see Playwright connected.
Using Playwright from Claude Code
Example instruction:
Using the Playwright MCP server, visit http://localhost
On first invocation:
- Claude Code requests permission for navigation
- Permission must be approved explicitly
- The session now controls a browser instance
Capabilities unlocked:
- Automatic page load detection
- Screenshot capture as context
- Console error inspection
- Visual diffing during UI iteration
- Scroll automation
This turns Claude Code into a functional testing partner.
Not theoretical.
Operational.
Permission model: explicit and session-scoped
Each MCP tool requires confirmation the first time it runs in a session.
For example:
- Navigate to URL
- Capture screenshot
- Access Notion workspace
- Query Linear issues
Permissions are:
- Tool-specific
- Session-scoped
- Explicitly approved
This design preserves control.
The model never acts silently.
Installing the Notion MCP Server
Notion MCP brings documentation into context.
Transport: typically HTTP.
Installation requires authentication.
Claude Code will:
- Open a browser window
- Ask you to authorize workspace access
- Confirm token linkage
Once connected, you can say:
Using the Notion MCP server, read the product specification document for Ratings V2.
Claude Code receives structured JSON:
- Title
- Content blocks
- Metadata
No copy-paste.
No manual context injection.
Documentation becomes first-class input.
Installing the Linear MCP Server
Linear MCP integrates project management into reasoning.
Transport: Server-Sent Events (SSE).
After installation and authentication:
You can request:
Using the Linear MCP server, list issues assigned to me in the Ratings milestone.
Claude Code retrieves:
- Issue titles
- Status
- IDs
- Descriptions
You can then:
- Create issues
- Update status
- Move tickets
- Generate task breakdowns
Now planning is connected to execution.
The professional workflow unlocked by MCP
With all three servers installed, a full-stack engineering flow emerges:
Step 1 — Bring product context
Use Notion MCP to fetch requirements.
Step 2 — Generate implementation plan
Claude Code synthesizes architecture from documentation.
Step 3 — Create execution tasks
Linear MCP generates structured tickets.
Step 4 — Implement feature
Claude Code assists in code generation and refactoring.
Step 5 — Validate UI
Playwright MCP navigates the application, detects visual issues, inspects console errors.
Step 6 — Iterate
Fix issues → Revalidate → Update Linear.
This is not theoretical orchestration.
This is operational integration.
Why MCP changes engineering culture
Without MCP, Claude Code is a reasoning tool.
With MCP, it becomes:
- A documentation-aware collaborator
- A browser-enabled tester
- A task-aware planner
- A cross-system analyst
The shift is subtle but profound:
You stop treating AI as a coding autocomplete.
You start treating it as infrastructure-aware cognition.
Security implications
MCP is powerful because it is explicit.
It does not:
- Bypass permissions silently
- Execute arbitrary external commands
- Access workspaces without authentication
Each server defines:
- Tool boundaries
- Input constraints
- Output format
- Transport channel
This keeps authority human-controlled.
Architectural perspective
From a systems viewpoint, MCP introduces:
- Externalized context channels
- Explicit capability negotiation
- Transport abstraction
- Tool-based action gating
It resembles a microservice architecture — but for model cognition.
Each MCP server is a bounded context.
Claude Code orchestrates them.
Final reflections
Installing MCP servers is not a configuration step.
It is a mindset shift.
You are designing the cognitive surface of your assistant.
When Playwright validates UI, Notion informs requirements, and Linear tracks execution — Claude Code becomes embedded in your workflow.
Not as a spectator.
But as a participant.
Next chapter: designing custom MCP servers and advanced tool orchestration.
— Cristian Sifuentes
Full-stack engineer · AI-assisted systems thinker
Top comments (0)