Listed from the MCP registry; nobody on Odel publishes it.
Kia Access for Claude — vehicle status, location, doors, climate, and charging
Runs on your machine from its package; the steps are below.
At most: moves money, runs code or commands, deletes data, or sends messages for you.
Odel charges nothing per call. The publisher may charge for its own service.
Runs locally over stdio
This server isn't hosted — your MCP client launches it from a package registry. Use one of the commands below, or drop the config into your client (e.g. Claude Desktop).
Run
npx -y kiaaccess-mcp
MCP client config
{
"mcpServers": {
"kiaaccess": {
"command": "npx",
"args": [
"-y",
"kiaaccess-mcp"
]
}
}
}A Model Context Protocol server that connects Claude to your own Kia vehicle through the Kia Owners API the Kia Access mobile app uses: vehicle status, location, odometer, EV charge state, and confirmation-gated door, climate, and charging commands.
Warning
AI-developed project. This codebase was built and is maintained by Claude Code. No human has audited the implementation. Review the code and the tool permissions before pointing it at a real car.
Caution
This server can move a two-tonne object and can unlock your car. Every command tool asks you to confirm before it acts — with a confirmation prompt where the client supports one, otherwise a preview plus a one-time token that only a repeat call can use (see Confirmations) — and door lock/unlock is not even registered unless you opt in with KIA_WRITE_MODE=all. Read Vehicle commands before changing that.
Ask Claude things like:
By using this server you accept that:
loginAttempt) and eventually sets enforceRecaptcha, after which server-side login for that account is impossible. This server therefore never retries a rejected credential — see If login fails.git clone https://github.com/chrischall/kiaaccess-mcp.git
cd kiaaccess-mcp
npm install
npm run build
cp .env.example .env
# Edit .env: KIA_USERNAME, KIA_PASSWORD (and optionally KIA_WRITE_MODE,
# KIA_DEVICE_ID, KIA_RMTOKEN, KIA_REQUEST_TIMEOUT_MS — see .env.example)
.env is gitignored. The server never logs credentials, and no tool ever returns your password.
{
"mcpServers": {
"kiaaccess": {
"command": "npx",
"args": ["-y", "kiaaccess-mcp"],
"env": {
"KIA_USERNAME": "you@example.com",
"KIA_PASSWORD": "your-password",
"KIA_WRITE_MODE": "comfort"
}
}
}
}
Missing credentials do not stop the server from booting — it starts, answers tools/list, and only reports the configuration error when a tool actually needs to call Kia. Run kia_session_status to see what it thinks it has.
Kia challenges every new device with a one-time passcode. This server bootstraps once, stores the resulting remember-me token (rmtoken) under ~/.kiaaccess-mcp/session.json, and from then on mints fresh sessions silently — Kia does not rotate the token and does not challenge again.
Run it through Claude, in this order:
kia_session_status — confirms credentials are present. If it reports hasSession: false, continue.kia_start_login (asks you to confirm first) — sends your credentials, returns an otpKey and an xid, plus the masked phone/email Kia has on file.kia_send_otp — pick SMS or EMAIL. The passcode expires in about two minutes.kia_verify_otp — hand it the passcode. The token is stored locally and is deliberately not returned.kia_list_vehicles — confirms the session works and gives you the vehicleKey every other tool takes (as its vehicle_key argument).To start over (revoked token, changed password, handing the machine on), run kia_forget_session (asks you to confirm first) and repeat from step 2.
The bootstrap needs a human once. A deployment that has no one to read an OTP — a remote host — cannot run it at all, so bootstrap on a machine that can and move the token:
KIA_DEVICE_ID to a fixed uuid before the bootstrap, on both machines. The rmtoken is minted against a device uuid and is worthless with a different one, and kia_session_status only ever reports a truncated prefix — so if you let it be generated, you cannot read back the value you need.rmtoken value out of ~/.kiaaccess-mcp/session.json yourself, in a terminal. No tool returns it: the token is a long-lived MFA bypass, so it is never put into a conversation.KIA_RMTOKEN, alongside KIA_USERNAME, KIA_PASSWORD and the same KIA_DEVICE_ID.KIA_RMTOKEN takes precedence over anything in the local session store, so the deployment's session is whatever you handed it rather than whatever it last wrote. Treat the value like the password it stands in for: it bypasses MFA, and with the account password it grants full control of the vehicle.
Do not retry. Kia increments loginAttempt on every rejection and eventually sets enforceRecaptcha, which breaks server-side login for that account permanently. Verify the email and password in the Kia Access app first, fix .env, restart, and only then try again.
KIA_WRITE_MODE decides which command tools are registered at all. This is a structural gate, not a runtime check: a tool that was never registered cannot be invoked by any host permission setting or by an instruction injected into the conversation.
KIA_WRITE_MODE | Registers |
|---|---|
none | Nothing but the read tools and the account tools |
comfort (default) | Climate start/stop and the charging commands |
all | Also kia_lock_doors and kia_unlock_doors |
An unrecognised value fails closed to none and warns on stderr — a typo must never silently grant the ability to unlock a car.
Two more rules hold for every command:
commandAccepted and stateConfirmed separately. Observed changes took 30–60 seconds.Every write tool (the commands above, plus kia_start_login and kia_forget_session) asks you to confirm before it acts. A client that can show a confirmation prompt (Claude Code) shows one, with the exact request. A client that cannot (claude.ai, Claude Desktop) gets a two-step flow instead: the first call does nothing and returns a preview plus a confirmToken, and only a repeat call carrying that token proceeds. The token is bound to the exact request previewed — change an argument and it is refused (DRAFT_CHANGED) with a fresh preview; use it twice and the second call is refused (TOKEN_REUSED).
| variable | default | |
|---|---|---|
MCP_CONFIRM_MODE | ask-user | What a write does on a client that cannot show a confirmation prompt (claude.ai, Claude Desktop). ask-user: two steps — the first call does nothing and returns a preview plus a token, and the model must get your approval in chat before calling again with it. auto: the same two steps, but the model may use the token after reviewing the preview itself. refuse: writes are refused on such clients. A client that can show prompts (Claude Code) always gets the real prompt, unless MCP_CONFIRM_ELICITATION=off. An unrecognised value is treated as refuse. |
MCP_CONFIRM_ELICITATION | on | off never shows a confirmation prompt, so every client gets the MCP_CONFIRM_MODE behaviour. Set it for a client that says it can show prompts but never does (the write hangs — opencode 2.0.x). Any other value is treated as on, with a warning on stderr. |
MCP_CONFIRM_TTL_SECONDS | 600 | How long a token stays valid. |
MCP_CONFIRM_SECRET | random per process | Signing key; set it only if tokens must survive a server restart. |
MCP_HOST_CONFIRM_SECRET | unset | Set by mcp-host for this registration (it has a data dir). Used when MCP_CONFIRM_SECRET is unset and MCP_DATA_DIR is absolute: a token survives the hosted child idling out between the preview and the confirmed call, and spent tokens are recorded under the data dir so a restart cannot re-accept one. |
| Tool | Notes |
|---|---|
kia_session_status | Configured? Bootstrapped? Which write mode? No network call, no secrets — the email is masked and the device id truncated. |
kia_start_login | Step 1 of the MFA bootstrap. Confirmation-gated, because a rejection has a permanent cost. |
kia_send_otp | Step 2 — delivers the passcode by SMS or EMAIL. |
kia_verify_otp | Step 3 — exchanges the passcode for a stored session. Returns no secret. |
kia_forget_session | Discards the stored token so the bootstrap can be re-run. Local only; confirmation-gated. |
| Tool | Notes |
|---|---|
kia_list_vehicles | Every enrolled vehicle with its vehicleKey, nickname, model, mileage. VINs are masked to the last 6 characters. |
kia_vehicle_status | Cached status: door lock, ignition (ign3 — on an EV engine stays false), the nested climate block, and more with include_raw. Fast, but only as fresh as the last upload. |
kia_refresh_status | Wakes the telematics unit for a fresh reading. Slower, draws a little power, and returns no data itself — read kia_vehicle_status afterwards. |
kia_vehicle_location | Last reported position plus a map link. Not a live GPS fix. |
kia_charge_targets | Target state of charge per plug type. |
| Tool | Mode | Notes |
|---|---|---|
kia_start_climate | comfort | Preconditioning. Temperature is best-effort: the car may report its own last-set target instead of the one requested. |
kia_stop_climate | comfort | Verified by re-reading climate.airCtrl. |
kia_start_charge | comfort | Verified. Needs the car plugged in — unplugged, Kia accepts the request and nothing happens. Confirm via evStatus.batteryCharge. |
kia_stop_charge | comfort | Verified. Confirm via evStatus.batteryCharge. |
kia_set_charge_limits | comfort | Verified, and re-read against evc/gts afterwards. Send both plug types — the list replaces the stored one. |
kia_lock_doors | all | Verified by re-reading doorLock. |
kia_unlock_doors | all | Leaves the car physically unsecured. Only run it when the user explicitly asked. |
Every endpoint here was verified live against a 2024 EV9 — the reads and the four door/climate commands on 2026-07-27, and the three evc/* charging commands on 2026-07-28 against a plugged-in car. Each was proven by re-reading state, never by the success status: Kia answers statusCode: 0 the moment it accepts a command, seconds before the car acts, and on an unplugged car it answers success and does nothing at all. Every command result therefore reports "accepted" and "confirmed" separately. The full protocol write-up is in docs/KIA-API.md.
npm test # unit tests (no network — fetch is mocked throughout)
npm run test:coverage # the same, with the enforced 100% thresholds
npm run build # tsc + esbuild bundle