MCP Integrations

Quick Start: Sequant works fully without any MCP servers, but optional integrations enhance browser testing, library documentation lookup, and complex reasoning capabilities.

Overview

MCP (Model Context Protocol) servers extend Claude’s capabilities with specialized tools. Sequant supports three optional MCP integrations that enhance specific workflows.

Looking for Sequant’s own MCP server? This guide covers the optional third-party MCPs below. Sequant also ships its own MCP server (sequant serve, tools sequant_status/sequant_run/sequant_logs), configured in .mcp.json. It’s pinned to your installed version rather than @latest so reconnects don’t reinstall on every release (#793) — sequant init writes the pin and sequant update refreshes it. See MCP Server → Version pinning, and Troubleshooting → MCP Server Issues if a reconnect fails with -32000.

Supported MCPs

MCP Server	Skills Enhanced	Purpose
Chrome DevTools	/test, /testgen, /loop	Browser automation for UI testing
Context7	/exec, /fullsolve	External library documentation lookup
Sequential Thinking	/fullsolve	Complex multi-step reasoning

Checking MCP Status

Run the doctor command to see which MCPs are configured:

npx sequant doctor

Example output:

✓ MCP Servers: Some MCPs configured: sequential-thinking
⚠ MCP: chrome-devtools: Not configured (optional, enhances /test, /testgen, /loop)
⚠ MCP: context7: Not configured (optional, enhances /exec, /fullsolve)

Note: These lines detect MCPs from your Claude Desktop config — a different source than the .mcp.json phase agents actually use for sequant run (#936), so a server present here may still be unavailable headlessly, and vice versa. Doctor’s separate “MCP Servers (headless)” line reports what a phase actually gets (sequant + .mcp.json + any mcpAllowlist names) — check that line for headless availability.

Installation

Add these to the project’s .mcp.json (the same file sequant init already writes a sequant entry into) — not your Claude Desktop config. sequant run phase agents read only .mcp.json and never pass through your Claude Desktop config (#936), so a server configured only in Claude Desktop is invisible to /test, /exec, and /fullsolve when they run headlessly.

Chrome DevTools MCP

Enables automated browser testing with snapshots, screenshots, and interaction recording.

1. Follow the setup guide for the exact command/args
2. Add the resulting entry to .mcp.json’s mcpServers object at the project root

Without this MCP: The /test skill falls back to generating manual test checklists instead of automated browser tests.

Context7 MCP

Provides up-to-date library documentation during implementation.

Add an entry to .mcp.json:

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp"]
    }
  }
}

Without this MCP: The /exec skill uses codebase search and general knowledge instead of fetching current library docs.

Sequential Thinking MCP

Enables extended reasoning for complex multi-step problems.

1. Follow the setup guide for the exact command/args
2. Add the resulting entry to .mcp.json’s mcpServers object at the project root

Without this MCP: The /fullsolve skill uses standard reasoning without extended thinking chains.

.mcp.json is committed to git — don’t put literal secrets in an entry’s env. An MCP that needs a credential isn’t a good fit for phase-agent use; keep it in your interactive Claude Code / Claude Desktop config instead.

If a server genuinely only exists in your Claude Desktop config — e.g. one you’re not ready to commit yet — settings.run.mcpAllowlist (.sequant/settings.json) is a deliberate, per-server opt-in to pass it through anyway. It’s a security-relevant decision, not a convenience default: see Allowlisting a Desktop-Only Server before using it.

Behavior Without MCPs

Sequant is designed to work without any MCPs. Here’s what changes:

Scenario	With MCP	Without MCP
/test command	Automated browser snapshots and interactions	Manual test checklist generated
/exec library lookup	Fetches current docs from Context7	Uses codebase patterns and general knowledge
/fullsolve reasoning	Extended thinking for complex problems	Standard reasoning approach

Troubleshooting

MCP not detected by doctor command

Symptoms: You installed an MCP but sequant doctor still shows it as not configured.

Solution:

1. Verify the MCP is in your Claude Desktop config file
2. Check the server name matches expected patterns (e.g., contains “chrome-devtools”, “context7”, or “sequential-thinking”)
3. Restart Claude Desktop after config changes

Skills fail with MCP errors

Symptoms: Error messages about missing MCP tools when running skills.

Solution:

1. Run sequant doctor to check MCP status
2. If MCP is optional for your use case, the skill should fall back gracefully
3. If you need MCP functionality, install following the guides above

Claude Desktop config location

Platform	Config Path
macOS	~/Library/Application Support/Claude/claude_desktop_config.json
Windows	%APPDATA%\Claude\claude_desktop_config.json
Linux	~/.config/claude/claude_desktop_config.json

---

Generated for Issue #15 on 2025-01-10