How to Connect Claude Code to fanTask via MCP: Step-by-Step Guide

What is the Model Context Protocol (MCP) and why does it matter?

The Model Context Protocol (MCP) is an open standard developed to allow Large Language Models and AI coding assistants to securely interact with external tools, databases, and services. Instead of confining an agent to the local files inside its immediate directory or relying on copy-pasting terminal output between separate browser tabs, MCP provides a structured, native protocol over encrypted JSON-RPC. This enables AI tools to query live data, execute specialized operations, and update external systems directly during their problem-solving loop.

Why connect Claude Code to fanTask?

Claude Code is a fast, capable terminal-based coding assistant. However, like all state-of-the-art LLMs, it operates under the strict constraints of finite context windows. When long terminal sessions get compacted, or when you start a fresh session tomorrow, Claude loses all recollection of yesterday's architectural decisions, test conventions, and debugging breakthroughs.

fanTask acts as a durable external project ledger and memory layer for autonomous coding agents. By connecting Claude Code to fanTask via MCP, your agent gains persistent superpowers:

• It reads your living project conventions and architectural boundaries (via the persistent AI Context field) before writing a single line of code.
• It searches past task histories and comments to inspect how strange edge cases and runtime bugs were solved previously.
• It picks up, updates, and documents tasks directly on a shared task board that human engineers and AI agents work from simultaneously.

Step 1: Get your access token from fanTask

If you just created your account, fanTask automatically shows you a "Connect your AI to fanTask" modal right on your workspace page, with a few options (Claude, Gemini CLI / Antigravity, ChatGPT, Gemini App, Other). Pick Claude — the platform mints a personal access token on the spot and hands you a ready-to-paste setup prompt for Claude Code or Claude Desktop.

If you closed that modal, or you're on an existing account, you can pick this back up anytime from your workspace menu: Workspace → MCP / AI. It offers the same options and the same ready-to-paste prompt, generated with your current token.

Step 2: Register the fanTask server in Claude Code

The simplest path is to paste the Step 1 prompt directly into a Claude Code conversation — it already contains the registration command with your token filled in, and Claude runs it for you. If you'd rather run it yourself in a terminal, this is the exact command:

claude mcp add fantask --transport http https://api.fantask.ro/mcp --header "Authorization: Bearer YOUR_FANTASK_API_TOKEN" -s user

Replace YOUR_FANTASK_API_TOKEN with the token from Step 1. This registers the fantask MCP server at the user scope (-s user, available across all your Claude Code projects), with the required authorization header, pointed at the production endpoint https://api.fantask.ro/mcp.

Step 3: What changes once connected

When you launch Claude Code in your workspace, the agent automatically initializes the fanTask MCP connection and registers the following suite of tools:

• fantask.get_project_context(project_id): Retrieves the living AI Context document for your project. This includes your established tech stack boundaries, gotchas (e.g. browser cookie limitations, database lock caveats), and required test commands.
• fantask.search_tasks(query, filters): Runs keyword and semantic queries across previous tasks, discussions, and resolution notes to avoid re-solving previously tackled bugs.
• fantask.get_task(task_id): Fetches full issue details, acceptance criteria, and audit comments for a specific ticket.
• fantask.update_task(task_id, status, comment): Transitions a task's status (such as moving from todo to in_progress or in_review) and records explanatory notes.
• fantask.create_task(project_id, title, description, priority): Automatically opens subtasks or follow-up tickets on the board when the agent discovers new technical debt or edge cases.
• fantask.add_comment(task_id, message): Appends clear, structured audit notes detailing files touched, test results, and rationale for future reference.

Recommended Collaborative Workflow

With fanTask connected, an effective session with Claude Code looks like this:

1. Session Initialization: You start by telling Claude: "Review our project AI Context in fanTask and check active tasks." Claude immediately executes get_project_context and search_tasks, grounding itself in your conventions before proposing modifications.
2. Active Implementation: Claude marks the target ticket as in_progress via update_task. If it encounters an ambiguous error or unfamiliar SDK behavior, it calls search_tasks to see if your team solved it before.
3. Review & Hand-off: Once the code is written and tests pass, Claude updates the ticket to in_review and posts a concise changelog using add_comment. You inspect the git diff and the fanTask web UI, then verify and mark the task as done yourself.

Get Started for Free

The fanTask Free plan gives you one complete project board with up to 3 team members, persistent AI Context, and full MCP read and write capability. Paid plans add additional seats and projects.

Create a free fanTask account and connect Claude Code today →