Support
Most connection problems are fixable in a minute or two with locus doctor and the setup guide below. If that doesn't get you unstuck, email us directly — a person reads every message.
Support today is this page, email, phone, and two MCP tools your AI assistant can call for you.
Your AI assistant can already check your account status and diagnose your connection directly, from inside the client you're already using — see "Support tools inside your AI assistant" below for both. Only the third tool, general help (ask a question, or reach a human), is still designed but not built. Until it ships, this page and email cover that gap.
Fix a connection problem
Start local, before assuming anything is wrong with your account or the setup steps:
locus doctorIt checks your Python version, every dependency, your configured embedder's API key, whether your data directory is writable, and does a real embed-and-store round trip — then tells you exactly which check failed and what command fixes it. It's the same first step we'd ask you to run if you emailed us.
Once locus doctor passes but your AI client still isn't finding Locus, the problem is almost always in that client's own MCP config — see Connect Locus to your AI assistant for exact, per-client steps (Claude Desktop, Claude Code, Cursor, Cline, GitHub Copilot, and more, each with its own config shape and known gotchas).
Common issues and fixes
`locus doctor` reports a FAIL line
Read the specific line — it names the exact check (Python version, a missing dependency, the embedder API key, or the data directory) and the exact command to fix it. Re-run locus doctor after each fix; it exits clean only once every check passes.
Your AI assistant doesn't list `locus_search` / `locus_list_sources` / `locus_read_file`
Almost always a config-file problem, not a server problem: an MCP config with a bare, relative or wrong path to the locus command, or a client that needs a full restart (not just a reload) to pick up a newly added server. Confirm the command is the absolute path that command -v locus prints, with serve as its argument, then check your client's specific setup steps — several clients use their own config file shape.
`locus query` / a search tool returns no results
Nothing is indexed yet, or you're querying the wrong collection. Run locus stats to see what's indexed, then locus index <folder> --prune to (re)index it.
A dependency fails to import, or the embedder/store check fails
Reinstall the engine the way you installed it: after a pipx install, pipx reinstall locusfiles; after a uv install, the same uv tool install command with --reinstall. From a source checkout (developers), re-run scripts/setup.sh local (or pip install -e ".[dev,local]") inside the activated virtual environment. These steps are for macOS and Linux; Windows is not yet verified. If you're using voyage or openai as your embedder, confirm the matching API key is actually set in .env, not just exported in a shell that isn't the one your MCP client launches.
You just installed a new AI client and it doesn't see any MCP servers
Some clients (Kilo Code and a few others) have changed their config file shape between versions, so a config that worked for an older release can silently stop being read. Check that client's entry for the current shape rather than reusing an old snippet from memory or a search result.
By AI assistant or client
Every client in the connect guide's setup list has its own step-by-step section and a label for how far we have tested it (Documented and Connects mean not yet verified). A few have known version-specific quirks worth reading before you copy a config from somewhere else:
- Goose — needs a configured LLM provider before it will even load an MCP extension, separate from the extension registration itself.
- Kimi CLI, Qwen Code CLI and DeepSeek Harness — Chinese-market clients; in our tests, connecting Locus needed no client-vendor API key, only a model-provider key if you also want that client's own chat to work.
- Kimi Code CLI (Kimi CLI's successor) and ZCode — not yet tested by us, so we cannot say yet whether connecting Locus needs a vendor key or sign-in. Follow the steps in each entry.
- Anything else — see the full list.
Account and usage
Your plan, trial status, daily usage, and API keys (create, view, revoke) are on your dashboard. That page shows the real, current numbers — nothing here duplicates or estimates them.
Lost or leaked an API key: revoke it from the dashboard and issue a new one — the same action whether it was compromised or just misplaced. Deleting your account and its data is also a dashboard action (Privacy → Your control).
You can also just ask your AI assistant
Before you email us: the assistant you're already talking to is generally good at this kind of troubleshooting on its own, with no Locus-specific tool required. Paste it locus doctor's output, or the relevant section of this page, and ask it to walk you through the fix — that's ordinary reasoning any capable model can do over troubleshooting text, not a feature Locus built or trained. For a config or connection problem it's often faster than waiting on an email reply. If you have a hosted account, your assistant can also check its real status directly — see below.
Support tools inside your AI assistant
locus_account_status is real and registered on the MCP server today — ask your assistant to check your Locus account and it calls this directly. It never takes a parameter naming an account: it only ever reports on whichever account your own LOCUS_API_KEY belongs to, and works without one configured too (it just says so, rather than erroring). A canceled or past-due account still gets a real, accurate status report — this tool never gates on subscription status, only reports it truthfully.
locus_diagnose_connection is also real — ask your assistant to diagnose your Locus connection and it reports your local setup (embedder, whether the right API key is present, data-directory status, indexed chunk count) with no secrets, file contents, or file paths ever included. Add "including remote reachability" and it also checks whether Locus's own hosted endpoints answer, without ever showing you what they respond with.
One more is specced but not yet built — locus_get_help (ask your assistant for docs, or reach a human). When it ships: no account required for basic help, and any support-ticket submission only ever happens after you explicitly confirm it in that turn — file or search content will never be able to trigger a ticket on its own.
Contact us
Still stuck, or something above doesn't cover it? Email yonilev2003@gmail.com — include the output of locus doctor and which AI client you're using; it's the single most useful thing you can paste in and usually skips a round trip of back-and-forth questions.
Billing, cancellations and anything else: yonilev2003@gmail.com or +972-54-619-2345 (Sunday–Thursday, 09:00–17:00 Israel time).
Optional. We use your message only to reply. Privacy · Terms · Sub-processors