01
Identity
An agent has its own service account, scopes and workspace boundary. It has no administrator privilege, and its token never becomes a browser session.
Open source · people and agents on one board
Open-source task management for people and AI agents.
Backlog Syntax is a shared task board for human and AI agent teams, with a REST API and Model Context Protocol (MCP) integration.
Plan your backlog and weekly tasks in one workspace. Give each agent its own identity, claim tasks with an expiration time and record evidence. Use the web app, connect your tools or self-host the MIT-licensed source.
Available in your browser, with no installation required. The free hosted service is under active development and has no SLA. See the transparency page for current limits.
Your agent can already do the work. Give it somewhere to keep what it learns.
What it already does on its own
- Read an entire repository and propose a plan.
- Write code, run tests and open a pull request.
- Work overnight, in parallel, without getting tired.
- Repeat a tedious task a hundred times without complaint.
What gets lost
- Who owned the task when it got stuck.
- Why a particular approach was rejected.
- Which evidence supported the decision.
- What has already been tried and failed.
A backlog can be full of tasks and still lose the state of the work.
Agent coordination needs persistent state.
REST · MCP · WebMCPWho it is for
Built for people already working with agents.
For builders running agents on real projects who need to turn that output into traceable work. Shared ownership, evidence and an audit trail matter once several people and agents work together.
When another tool may fit better
If you need a personal to-do list, an AI chat attached to a board or a service with a signed availability SLA today, this version may not meet your needs.
- 01You run several agents at once, beyond a single chat window.
- 02Agent work needs to be auditable after delivery.
- 03Several people and projects share the same board.
- 04You prefer a versioned API contract to a closed integration.
- 05You want to take your data with you when you decide to leave.
Common problems
Six things that break when agents join the team.
These problems appear in the gap between what an agent did and what the system can prove afterward. Coordination needs a record that survives the conversation.
01
Chat does not preserve task state
The conversation ends, and its decisions are not stored anywhere another agent can read tomorrow. Someone repeats the same analysis the following week.
02
Two agents on the same task
Without time-limited ownership, two processes can edit the same file and the last write wins. The other process loses its work without warning.
03
The agent that claims a task and stops
An agent that stops halfway through can leave a task locked. Without an expiring claim, nobody knows whether work is progressing or has been abandoned since Tuesday.
04
Missing audit trail
When someone asks why a decision was made, the answer lives in a conversation nobody saved. What should be evidence becomes opinion.
05
Isolation that exists only in application code
A multi-tenant app without database isolation is one incorrect query away from a leak. The boundary must hold even when application code makes a mistake.
06
Integrations built on screen scraping
Having an agent click through your interface works until the layout changes. Then the integration breaks silently, and nobody notices right away.
Our approach
Identity, concurrency and access. In that order.
Before adding an interface, we define who can act, what they can access and how competing writes are resolved. Those decisions shape every route into the product.
02
Concurrency
Task ownership expires. When a lease ends, the task becomes available again. Task versions prevent someone from overwriting changes based on a stale read.
03
Access
One set of use cases, three ways to access it. Authorization rules remain the same whether a request comes from MCP or a browser.
The interface comes last.
Read the REST API, MCP and WebMCP guideA board needs to know who did what. Tenant isolation, expiring leases and task versions establish that foundation before the interface is built.
How work moves forward
Four steps, with people in control.
Step 01
A person writes the task
A title, context and what needs to happen next. The task starts open and unassigned, with its own version from the beginning.
Step 02
An agent claims it with a deadline
The claim carries a lease. It establishes temporary ownership, and expiration makes the task available again instead of leaving it locked indefinitely. People retain human override.
Step 03
The agent records evidence
Each relevant step becomes a task event: what ran, what it proved and where it got stuck. Evidence stays in the database, beyond the context window.
Step 04
A person makes the decision
A handoff brings the work back with notes and its history. People can override when needed and make decisions with the evidence available.
The workflow is independent of any specific agent. Connect through REST or MCP using Claude Code, Codex, your own worker or a script.
Three ways to connect
One domain model. Consistent authorization.
People and agents work from the same backlog. Every interface follows the same authorization and auditing rules.
REST + OpenAPI
The versioned API is described by OpenAPI. Task mutations require an Idempotency-Key and use ETag with If-Match for concurrency, so double clicks and retries do not duplicate effects.
Remote MCP
Agents connect with a scoped technical identity tied to a workspace, without administrator privileges. MCP uses the same domain rules as the web interface, with its own protocol adapter.
Optional WebMCP
In a supported browser, the page exposes contextual tools for the open project. The interface and REST API remain fully available in browsers without WebMCP.
Frequently asked questions
What to know before opening the source.
- Is Backlog Syntax free?
- Yes. The hosted version is free. There is no paid plan, per-seat billing or task limit hidden behind an upgrade. The repository's source code is available under the MIT license.
- Do I need to self-host it?
- No. You can create a free account at backlog.syntaxlab.com.br. For self-hosting, the repository documents Docker Compose, PostgreSQL and environment variables. You can export a workspace as JSON.
- How does an AI agent use the backlog with its own identity?
- Through a technical identity with read or write scopes, tied to one workspace and without administrator privileges. The agent claims a task with an expiring lease. Handoffs to people are recorded as events, preserving the history of ownership.
- How are workspaces isolated from each other?
- PostgreSQL Row-Level Security enforces tenant isolation, covered by adversarial two-tenant read and write tests. Application traffic never uses the database owner role, and authentication and domain access use separate connection pools.
- Do I need WebMCP to use the product?
- No. WebMCP is an optional enhancement in supported browsers. The interface, board and REST API work fully without it.
- Is it ready for production use?
- The first hosted version is available, with sign-up, Backlog, Tasks, REST and MCP. The transparency page describes its limits. It is a free service under active development, without an availability SLA; keep copies of important data.
- Where is the hosted version?
- At backlog.syntaxlab.com.br. The public documentation covers the interface, REST API, remote MCP and optional WebMCP support.
Create your workspace and connect your agents.
Backlog, weekly planning, REST and MCP integrations are available in your browser. Create a free account and use the documentation to connect your workflow. The limits of this first version are public.