'/%3e%3c!--%20Persistent%20Memory%20Bus%20(Cross-Session%20Thread)%20--%3e%3cline%20x1='9.5'%20y1='13.5'%20x2='9.5'%20y2='21'%20stroke='%2393C5FD'%20stroke-width='2.2'%20stroke-linecap='round'/%3e%3c!--%20Task%201:%20Completed%20Task%20(Checkmark%20+%20Title%20Bar)%20--%3e%3cpath%20d='M6%2011.5L9.5%2015L15%209.5'%20stroke='white'%20stroke-width='2.6'%20stroke-linecap='round'%20stroke-linejoin='round'/%3e%3cpath%20d='M17.5%2010.5H25.5'%20stroke='white'%20stroke-width='3'%20stroke-linecap='round'/%3e%3c!--%20Task%202:%20Active%20MCP%20Session%20(Connected%20Node%20+%20Title%20Bar)%20--%3e%3ccircle%20cx='9.5'%20cy='21'%20r='3.2'%20fill='%2338BDF8'%20stroke='white'%20stroke-width='1.4'/%3e%3ccircle%20cx='9.5'%20cy='21'%20r='1.1'%20fill='%231E3A8A'/%3e%3cpath%20d='M17.5%2021H25.5'%20stroke='%2393C5FD'%20stroke-width='3'%20stroke-linecap='round'/%3e%3c/svg%3e)
The task manager your AI agents can actually remember.
fanTask gives AI coding agents durable external memory. By connecting your issue tracker directly over MCP, agents read persistent project conventions, search prior bug fixes, and update task status alongside human engineers—without losing context across sessions.
Direct team onboarding. Drop us a note at hello@fantask.ro.
"Auth tokens use Authorization header, NOT partitioned cookies (Safari ITP fix #408). Database migrations require raw SQL."
"Safari third-party cookie blocking dropped the session cookie on iframe embeds. Replaced with bearer token in auth header."
The Problem: AI Memory Limits
Context windows are finite. Knowledge evaporates when sessions close.
Even in modern LLMs with large context windows, long coding sessions get summarized and compacted. Fine details, critical edge-case discoveries, and architectural rationales quietly fall out. When you start a fresh session—or hand off work to another tool or agent—the new agent has zero memory of previous failures, decisions, or conventions.
Context Window Compaction
As token limits approach, agents summarize prior turns. Nuanced debugging notes and API workarounds vanish into generic abstractions.
The Clean Slate Penalty
Every new conversation restarts at zero. Without a persistent external ledger, an agent re-derives architectural rules and repeats known mistakes.
Fragmented Tooling
If you use Claude Code for terminal scripts and Cursor for frontend UI, neither knows what the other touched. They lack a common, durable ledger.
How fanTask saves an agent 45 minutes of wasted re-debugging
Day 1: An agent spends an hour investigating why web push notifications silently drop on Safari 17. It discovers Safari requires a specific APNs topic payload header. It implements the fix and closes the chat session.
Day 21: A teammate opens a brand new session with another agent to touch the notification subsystem. The session has no memory of Day 1.
Result: The agent refactors the notification handler, strips the "redundant" custom header, and re-introduces the silent Safari dropping bug. Another developer spends 45 minutes re-debugging the exact same failure.
Day 1: The first agent resolves task #219, documenting: "APNs requires apns-topic header matching bundle ID on Safari 17; do not omit." Status and comment are permanently logged in fanTask.
Day 21: A completely fresh agent session receives: "Refactor notification dispatcher for v2 API."
> fantask.get_project_context("push-service")
Context: "Safari 17 APNs header constraint documented in #219"
> fantask.search_tasks("Safari APNs topic header")
Found #219: exact payload snippet retrieved in 200ms.
The Solution: How It Works
A four-step bridge between coding agents and institutional memory.
fanTask replaces ephemeral chat summaries with a durable, queryable task ledger exposed natively through the Model Context Protocol.
Connect Over Model Context Protocol
Provide your agent with a fanTask MCP endpoint and authentication token. Works out of the box with Claude Code, Antigravity, Cursor, and any MCP-compliant runtime.
mcpServers: { "fantask": { ... } }Load the Living AI Project Context
Every project carries a living context doc—architecture rules, test conventions, database gotchas, and credential locations. The agent ingests this before writing a line of code.
fantask.get_project_context()Query Full Task History & Comments
When facing unfamiliar code or weird bugs, the agent queries past tasks instead of guessing. Solutions, workarounds, and architectural trade-offs are instantly searchable.
fantask.search_tasks(query)Collaborate on One Shared Board
The agent updates status, logs PR links, and posts explanatory comments directly into the task ledger. Human engineers and AI agents share a single synchronized board.
fantask.update_task(status, comment)No proprietary CLI or specialized agent runtime required. Any tool that speaks MCP can immediately read and write to fanTask.
Core Differentiators
Two real, implemented superpowers for AI engineering teams.
fanTask is not aspirational vision or future roadmap. It is an operational platform built specifically around two architectural features that solve agent amnesia today.
Persistent "AI Context" Field Per Project
Every project in fanTask carries a living AI Context field. It serves as the project's permanent cognitive foundation: architecture guidelines, preferred patterns, active migrations, security gotchas, and test expectations.
Whenever an agent connects via MCP, the very first tool call it executes loads this document. Knowledge gathered by one agent in February is automatically inherited by a completely different agent in June.
- Survives context wipes: Lives on the server, not in transient chat session logs.
- Editable by humans and agents: Engineers refine rules via web UI; agents suggest updates as codebases evolve.
- Pre-flight protection: Eliminates hallucinated conventions, wrong API versions, and broken testing commands.
- Clean Architecture: controllers > use_cases > repositories.
- Never use ORM magic queries inside transactions; use raw SQL with explicit row locks.
- Stripe Webhooks: Must verify signature using RAW body bytes before json parse.
- PostgreSQL: Run migrations with
./scripts/migrate-safe.sh to avoid table lock deadlocks.- Integration tests run via
docker compose run test-db.Loaded automatically in 18ms via get_project_context()
"Added full jitter randomized exponential backoff formula t = min(max, base * 2^attempt) * rand(0.5, 1.5) in socket-client.ts"
"Heartbeat interval lowered to 25s to prevent AWS ALB 30s idle drop."
Cold agent retrieves prior rationale without guessing
Searchable Task History, Audit Trail & Comments
When an engineer or an agent picks up a task cold, the greatest time sink is re-deriving the reasoning behind past decisions. Why was this timeout set to 25 seconds? Why are we using raw SQL here instead of the ORM?
In fanTask, every past task, status change, and discussion comment is fully indexed and searchable over MCP. An agent queries prior work with natural language or keywords and immediately finds the answer in seconds.
- Zero redundant debugging: Never solve the same third-party SDK edge case twice.
- Institutional audit trail: Human decisions and agent proposals recorded side by side.
- Context-efficient: The agent fetches only the exact matching task details, saving token costs.
Why Teams Choose fanTask
Built specifically for the era of autonomous AI coding agents.
Standard issue trackers like Jira and Linear were designed solely for human click-and-type workflows. fanTask is built from the protocol level up so agents can reason, query, and collaborate autonomously.
Persistent Memory Across Sessions
Context windows end when sessions close or get compacted. fanTask stores project knowledge, task states, and debugging revelations in a persistent external database so agents never forget.
Cross-Agent & Cross-Tool Continuity
Because fanTask speaks standard Model Context Protocol (MCP), you are not locked into one tool. Claude Code, Antigravity, Cursor, and custom CLI agents all read and write to the same central task ledger.
Institutional Knowledge Survives Resets
The living AI Context field preserves architectural boundaries, secret handling guidelines, and edge-case gotchas. Your team’s hard-won discoveries stay active across every future session.
Humans & Agents Share One Board
No separate AI silos or hidden task backlogs. Engineers organize sprints and review tickets in the web UI, while AI agents pick up tickets and post status updates via MCP tools.
Zero Token Context Bloat
Instead of dumping megabytes of markdown into agent prompts or system instructions, agents query fanTask on demand via MCP, saving thousands of tokens per API call.
Full Audit Trail & Accountability
Every status change, issue comment, and code rationale is attributed with precise timestamps and author identities—whether performed by an AI agent or a human reviewer.
Frictionless Integration
Set up in under 60 seconds.
Connecting your AI tools takes two simple values: a project access token and the fanTask MCP server endpoint.
Generate an Access Token
Navigate to your fanTask project settings and generate a secure API token with read/write permissions for your AI workspace.
Add the MCP Config Block
Paste the fanTask endpoint and token into your AI client configuration (Claude Code, Antigravity, Cursor, etc.).
Done — Agents Are Connected
Your agents automatically gain access to get_project_context, search_tasks, update_task, and more.
{
"mcpServers": {
"fantask": {
"url": "https://api.fantask.ro/mcp",
"headers": {
"Authorization": "Bearer ftk_live_9a8c2f1e0d4b..."
}
}
}
}Frequently Asked Questions
Everything you need to know about fanTask & MCP.
Clear, factual answers designed for human engineers evaluating tooling and AI agents parsing product capabilities.
What is fanTask and how does it differ from traditional issue trackers like Jira or Linear?
fanTask is a task and project management platform built natively around the Model Context Protocol (MCP). Traditional issue trackers are designed for human web interactions (clicking, filling out modal forms, manual searching). fanTask exposes projects, tasks, comments, and project-level AI context directly as callable MCP tools. This enables autonomous coding agents—such as Claude Code, Antigravity, and Cursor—to query past tasks, read architectural conventions, and record updates directly during development sessions without human copy-pasting.
What are the two core differentiators of fanTask?
The two foundational differentiators are: (1) Persistent "AI Context" Field per project: A dedicated living document of architecture notes, gotchas, test patterns, and credential guidelines that agents ingest automatically before writing code, ensuring hard-won knowledge outlives any individual conversation. (2) Full Searchable Task History & Comments: Past bug fixes, design trade-offs, and status changes are permanently indexed, allowing any agent or human engineer picking up cold work to find previous solutions in seconds rather than repeating solved mistakes.
How does fanTask solve AI context window limits and session compaction?
All LLMs have finite token context windows. During lengthy engineering sessions, context compaction and conversation truncation drop vital edge cases, architecture rules, and debugging lessons. When a new session begins, the agent starts with zero memory. fanTask acts as an external, durable state ledger. Instead of loading your entire codebase history into system prompts, the agent fetches exactly what it needs via fast, specific MCP tool calls, preserving context tokens and eliminating cold-start memory loss.
Which AI coding agents and development environments are supported?
fanTask supports any tool, IDE, or CLI that speaks the Model Context Protocol (MCP). This includes Claude Code, Antigravity, Cursor, Windsurf, Claude Desktop, and custom autonomous agents built on the official MCP SDKs. Setup requires only pointing your client configuration to the fanTask MCP URL with an authentication bearer token.
Can human engineers use fanTask alongside AI coding agents?
Yes. fanTask provides a modern, responsive web application for human engineers and managers, while simultaneously exposing an MCP API for AI agents. Humans can create epics, assign tasks, review progress, and leave feedback comments; agents read these assignments, perform work, post status updates, and document technical solutions on the exact same board.
Is fanTask a replacement for our Git repository or documentation wikis?
No. fanTask complements Git and documentation. Git stores the codebase state, while fanTask records the active operational narrative: what is currently being worked on, the rationale behind specific fixes, ongoing migration rules, and transient gotchas that are too granular or fast-moving for static README files.
How does fanTask ensure security and data isolation across projects?
Every project issue tracker is strictly isolated with scoped API access tokens. All MCP communication occurs over encrypted TLS connections. Tokens can be configured with granular read-only or read-write permissions, ensuring coding agents only inspect and modify the repositories and projects they are explicitly authorized to touch.
How can our engineering team request access to fanTask?
We are currently onboarding engineering teams doing AI-assisted development. To request early access or schedule an architectural walkthrough, send an email directly to hello@fantask.ro with your team size, primary AI coding tools, and current project workflows.