TeamCraft has no release yet. Security fixes go into the main branch
only; there are no maintained release branches.
| Version | Supported |
|---|---|
main (0.1.x, pre-release) |
yes – fixes land here |
| older commits, forks, snapshots | no |
TeamCraft has not had an independent security review. Until there is a stable release:
- run it only on trusted networks, or behind a reverse proxy with TLS;
- keep it up to date with
main; - keep backups of the database and the vault, as described in Operations.
Please do not open a public issue for security problems. Report them privately on GitHub: Report a vulnerability (Security tab → Advisories → Report a vulnerability).
Please include:
- the affected part: web UI, REST API, desktop app, file parsers or packaging;
- the commit you tested (
git rev-parse --short HEAD); - steps to reproduce or a proof of concept;
- the impact you expect.
Do not include real secrets or other people's data.
TeamCraft is a volunteer project, so there is no guaranteed response time. Every report is answered as soon as possible. We agree on a fix and a disclosure date with you, and credit you in the advisory unless you prefer otherwise.
In scope:
- authentication: passwords, sessions, API tokens, OpenID Connect – including a back-channel logout token that TeamCraft accepts without the provider's signature, or that ends another user's sessions;
- permission checks: roles, project groups, check-out locks, released data;
- read visibility of restricted project groups: any way for a non-member to read an item, file, preview, BOM line, change, request, notification, webhook payload or assistant answer of a restricted group, or to tell a hidden object from a missing one beyond the residual oracles listed there (unique numbers and codes, the shared number counter, where-used counts, response times);
- CSRF, XSS and the Content Security Policy;
- uploads and the parsers for CAD, mesh, point-cloud and BOM table files. This includes files that crash or hang the server, or that get around the time limit of the preview process;
- path traversal in the vault or in the desktop workspace;
- event webhooks that reach a target the target guard should refuse, or signatures that can be forged without the secret.
Out of scope:
- attacks that need administrator rights, or access to the server's file system or database;
- deployments that ignore the documentation, for example ones that expose the server without TLS;
- denial of service by plain request flooding;
- restricted data that a member already read: files copied to a desktop working folder stay there after access ends, audit entries from before migration 0024 are shown as written, backups and the command line contain everything, and the assistant sends what a member reads to the configured model provider.
- what a user's own AI agent does with their token: a local agent (the
desktop app's Agent view, an MCP client set up by hand) sends what it
reads to its own provider.
TEAMCRAFT_LOCAL_AGENTS=offcloses the MCP endpoint for every token and the desktop's Agent view then runs on the server-side assistant, so only that remains; the REST API stays reachable with a user's own token; anmcp_onlytoken cannot reach it and aread_onlytoken can only read through it (see Local agents). A call toPOST /api/v1/mcpthat gets through while the setting isoffis in scope. The desktop app answers a local agent's permission requests for TeamCraft's read-only MCP tools without a prompt (the server checks the user's rights on every call); a request it allows that is not a read-only tool of the TeamCraft server it registered – another server's tool, a write tool, the agent's own file or terminal tool – is in scope (desktop/src/autoallow.rs).