Perfect for Automated Workflow Agents needing seamless MCP server connections without context bloat. mcp-client is a universal client for connecting to MCP servers, allowing for efficient browser session management and multi-step operations.

How do I install mcp-client?

Run the command: npx killer-skills add cristian-robert/autonomous-automation-tester/mcp-client. It works with Cursor, Windsurf, VS Code, Claude Code, and 19+ other IDEs.

What are the use cases for mcp-client?

Key use cases include: Automating MCP server interactions, Generating browser sessions for testing, Debugging MCP call issues with browser_run_code.

Which IDEs are compatible with mcp-client?

This skill is compatible with Cursor, Windsurf, VS Code, Trae, Claude Code, OpenClaw, Aider, Codex, OpenCode, Goose, Cline, Roo Code, Kiro, Augment Code, Continue, GitHub Copilot, Sourcegraph Cody, and Amazon Q Developer. Use the Killer-Skills CLI for universal one-command installation.

Are there any limitations for mcp-client?

Each MCP call creates a new browser session. Browser closes after each call. Requires redoing all steps for returning to a previous state.

Universal MCP Client

Name: mcp-client
Availability: InStock
Author: cristian-robert

Connect to any MCP server without bloating context with tool definitions.

⚠️ PLAYWRIGHT USERS: READ "CRITICAL: Playwright Browser Session Behavior" SECTION BELOW!

Each MCP call = new browser session. Browser CLOSES after each call. You CANNOT navigate in one call and click in another. Use browser_run_code for ANY multi-step operation. If you need to return to a state (e.g., logged in), you MUST redo ALL steps from scratch.

How It Works

Instead of loading all MCP tool schemas into context, this client:

Lists available servers from config
Queries tool schemas on-demand
Executes tools with JSON arguments

Configuration

Config location priority:

MCP_CONFIG_PATH environment variable
.claude/skills/mcp-client/references/mcp-config.json
.mcp.json in current directory
~/.claude.json

Commands

bash
1# List configured servers
2python scripts/mcp_client.py servers
3
4# List tools from a specific server
5python scripts/mcp_client.py tools playwright
6
7# Call a tool
8python scripts/mcp_client.py call playwright browser_navigate '{"url": "https://example.com"}'

CRITICAL: Playwright Browser Session Behavior

⚠️ The Session Problem

Each MCP call creates a NEW browser session. The browser CLOSES after each call.

This means:

bash
1# ❌ WRONG - These run in SEPARATE browser sessions!
2python scripts/mcp_client.py call playwright browser_navigate '{"url": "https://example.com"}'
3python scripts/mcp_client.py call playwright browser_click '{"element": "Accept cookies"}'
4python scripts/mcp_client.py call playwright browser_snapshot '{}'
5# ^ The snapshot captures a FRESH page, not the page after clicking!

✅ The Solution: `browser_run_code`

Use browser_run_code to run multiple Playwright steps in ONE browser session:

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://example.com\");
4
5    // Wait for and click cookie banner
6    const acceptBtn = page.getByRole(\"button\", { name: /accept/i });
7    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
8      await acceptBtn.click();
9      await page.waitForTimeout(1000);
10    }
11
12    // Wait for page to stabilize
13    await page.waitForLoadState(\"networkidle\");
14
15    // Return snapshot data for analysis
16    const snapshot = await page.accessibility.snapshot();
17    return JSON.stringify(snapshot, null, 2);
18  "
19}'

When to Use Each Approach

Scenario	Tool	Why
Simple page load + snapshot	`browser_navigate`	Returns snapshot automatically
Multi-step interaction	`browser_run_code`	Keeps session alive
Click then observe result	`browser_run_code`	Session persists
Fill form and submit	`browser_run_code`	Session persists
Hover to reveal menu	`browser_run_code`	Session persists

Playwright Workflows for Test Discovery

1. Basic Page Exploration (Single Step)

browser_navigate returns both navigation result AND accessibility snapshot:

bash
1python scripts/mcp_client.py call playwright browser_navigate '{"url": "https://example.com"}'

Output includes:

Page URL and title
Full accessibility tree (all visible elements with roles, names, states)
Element references for further interaction

Use this when: Simple page load without interactions

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://www.olx.ro\");
4
5    // Handle cookie consent
6    try {
7      const cookieBtn = page.getByRole(\"button\", { name: \"Accept\" });
8      await cookieBtn.click({ timeout: 5000 });
9      await page.waitForTimeout(1000);
10    } catch (e) {
11      // No cookie banner
12    }
13
14    // Get accessibility snapshot
15    const snapshot = await page.accessibility.snapshot({ interestingOnly: false });
16    return JSON.stringify(snapshot, null, 2);
17  "
18}'

3. Navigate to Subpage (Multi-Step)

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://www.olx.ro\");
4
5    // Dismiss cookies
6    const acceptBtn = page.getByRole(\"button\", { name: \"Accept\" });
7    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
8      await acceptBtn.click();
9      await page.waitForTimeout(500);
10    }
11
12    // Navigate to login
13    await page.goto(\"https://www.olx.ro/cont/\");
14
15    // Wait for redirect to login domain
16    await page.waitForURL(/login\\.olx\\.ro/, { timeout: 10000 });
17
18    // Get form structure
19    const snapshot = await page.accessibility.snapshot();
20    return JSON.stringify({ url: page.url(), snapshot }, null, 2);
21  "
22}'

4. Explore Element Interactions (Multi-Step)

Use this to understand how menus/dropdowns behave:

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://www.olx.ro\");
4
5    // Dismiss cookies
6    const acceptBtn = page.getByRole(\"button\", { name: \"Accept\" });
7    if (await acceptBtn.isVisible({ timeout: 3000 }).catch(() => false)) {
8      await acceptBtn.click();
9    }
10
11    // Click on category to see what happens
12    const categoryLink = page.getByRole(\"link\", { name: /Auto, moto/i }).first();
13    await categoryLink.click();
14
15    // Wait to see result
16    await page.waitForTimeout(1500);
17
18    // Capture state after click
19    const snapshot = await page.accessibility.snapshot();
20    return JSON.stringify({
21      url: page.url(),
22      didNavigate: page.url().includes(\"auto\"),
23      snapshot: snapshot
24    }, null, 2);
25  "
26}'

5. Fill Form and Capture State

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://login.olx.ro\");
4
5    // Fill login form
6    await page.locator(\"input[type=email]\").fill(\"test@example.com\");
7    await page.locator(\"input[type=password]\").fill(\"test123\");
8
9    // Click login button
10    await page.getByTestId(\"login-submit-button\").click();
11
12    // Wait for response
13    await page.waitForTimeout(3000);
14
15    // Capture any error messages
16    const errors = await page.locator(\"[class*=error], [role=alert]\").allTextContents();
17    const snapshot = await page.accessibility.snapshot();
18
19    return JSON.stringify({
20      url: page.url(),
21      errors: errors,
22      snapshot: snapshot
23    }, null, 2);
24  "
25}'

Gathering Selectors for Page Objects

Best Practices

1. Use Accessibility Tree First

The snapshot from browser_navigate or browser_run_code provides:

Role: button, link, textbox, combobox, etc.
Name: accessible name (from label, aria-label, text content)
State: disabled, checked, expanded, etc.

Map these to Playwright locators:

typescript
1// From snapshot: { role: "button", name: "Căutare" }
2page.getByRole('button', { name: /Căutare/i })
3
4// From snapshot: { role: "textbox", name: "Ce anume cauți?" }
5page.getByRole('textbox', { name: /Ce anume cauți/i })
6
7// From snapshot: { role: "link", name: "Auto, moto și ambarcațiuni" }
8page.getByRole('link', { name: /Auto, moto/i })

2. Selector Priority

Priority	Method	Use When
1	`getByRole()`	Element has semantic role + accessible name
2	`getByTestId()`	Element has `data-testid` attribute
3	`getByText()`	Unique text content
4	`getByPlaceholder()`	Input with placeholder
5	`locator('[attr="value"]')`	CSS attribute selector
6	`locator('.class')`	CSS class (fragile, avoid)

3. Handling Multiple Matches

typescript
1// Use .first() when multiple match
2page.getByRole('link', { name: 'Category' }).first()
3
4// Use parent context
5page.locator('nav').getByRole('link', { name: 'Category' })
6
7// Use filter
8page.getByRole('button').filter({ hasText: /submit/i })

4. Get Full DOM for Complex Cases

When accessibility tree isn't enough, get raw HTML:

bash
1python scripts/mcp_client.py call playwright browser_run_code '{
2  "code": "
3    await page.goto(\"https://example.com\");
4
5    // Get specific element HTML
6    const formHtml = await page.locator(\"form\").first().innerHTML();
7
8    // Or get all buttons with their attributes
9    const buttons = await page.locator(\"button\").evaluateAll(btns =>
10      btns.map(b => ({
11        text: b.textContent,
12        testid: b.dataset.testid,
13        class: b.className,
14        type: b.type
15      }))
16    );
17
18    return JSON.stringify({ formHtml, buttons }, null, 2);
19  "
20}'

Quick Reference: Playwright MCP Tools

Tool	Session Behavior	Use Case
`browser_navigate`	New session, returns snapshot	Simple page load
`browser_run_code`	Single session, custom script	Multi-step operations
`browser_click`	New session	Single click (usually not useful alone)
`browser_type`	New session	Single type (usually not useful alone)
`browser_snapshot`	Reuses if session exists	Get current page state
`browser_screenshot`	Reuses if session exists	Visual capture

Tool Arguments

browser_navigate

json
1{"url": "https://example.com"}

browser_run_code

json
1{
2  "code": "await page.goto('https://example.com'); return await page.title();"
3}

The code must be valid JavaScript that:

Uses page object (Playwright Page)
Uses await for async operations
Returns the data you want (use JSON.stringify for objects)

browser_click

json
1{"element": "Submit button", "ref": "optional-element-ref"}

browser_type

json
1{"element": "Email input", "text": "user@example.com"}

Error Handling

Error	Cause	Fix
"No MCP config found"	Missing config file	Create mcp-config.json
"Server not found"	Server not in config	Add server to config
"Connection failed"	Server not running	Start the MCP server
"Invalid JSON"	Bad tool arguments	Check argument format
"Timeout"	Page too slow	Increase timeout in code
"Element not found"	Wrong selector	Check snapshot for actual names

Setup

Copy the example config:

bash
1cp .claude/skills/mcp-client/references/mcp-config.example.json \
2   .claude/skills/mcp-client/references/mcp-config.json

The config should contain:

json
1{
2  "mcpServers": {
3    "playwright": {
4      "command": "npx",
5      "args": ["@playwright/mcp@latest"]
6    }
7  }
8}

Install dependencies:
```
bash
1pip install mcp fastmcp
```

Config Example

See references/mcp-config.example.json

Available Servers

See references/mcp-servers.md for:

Playwright (browser automation)
GitHub (repository operations)
Filesystem (file access)
Sequential Thinking (reasoning)
And more...

Dependencies

bash
1pip install mcp fastmcp

mcp-client — mcp-client install mcp-client, autonomous-automation-tester, community, mcp-client install, ide skills, mcp-client browser session management, mcp-client multi-step operations, browser_run_code command, Claude Code, Cursor, Windsurf

About this Skill

Features

# Core Topics

Agent Capability Analysis

Ideal Agent Persona

Core Value

↓ Capabilities Granted for mcp-client

! Prerequisites & Limits

Browser Sandbox Environment

⚡️ Ready to unleash?

mcp-client