What you need from your partner
Two things, and they come from them, not from us:
Token A is not something you generate for this direction. The OCPI specification is
explicit that it is created by the receiving party and passed to you “in a secure way that
is outside the scope of this protocol”: an email, a portal, a phone call. If your partner
asks you for Token A, they mean to register into you instead, which is the opposite
direction and has its own page.
The address has no shape you can guess
The base URL travels out of band with Token A, and for the same reason: the spec never defines what it looks like. “The protocol is designed such that the exact URLs of the endpoints can be defined by each party”, and the example URLs in the specification “are descriptive only”. So/ocpi, /api/ocpi and /ocpi-2.2.1 are all legal bases, and only your partner knows which one
is theirs. Ask them for it rather than deriving it from anything else they sent you.
The one that catches people is the versions URL itself. What a partner hands over is usually the
full .../ocpi/versions, and this field wants the URL that hangs off, one segment shorter. Paste
the whole thing and we ask for .../ocpi/versions/versions, which no server advertised. On the
Endpoints page the line under the address names the exact URL the next press will ask for, so a
value one segment out shows itself before it costs you a connection. If you are not sure which of
the two you were given, press Check this address and we try the likely candidates and report
what each one answered.
Create the connection
You get back
201 with the connection, its _id, and status: "pending". Token fields
are never returned, by us or in any report.
Pick the version they actually run, not the newest one
The most common setup mistake is selecting 2.2.1 because it sounds current when the partner is on 2.1.1. Every subsequent failure will then be a version mismatch wearing the costume of a protocol bug. If you do not know what they run, ask, or read their versions endpoint. OCPI 2.1.1 is still in wide production use, and there is no shame in it.The credentials token is transported differently across versions. OCPI 2.1.1 sends it
raw; from 2.2 onwards it is Base64 encoded. Get the version wrong and authentication
fails on the very first request, with an error that looks nothing like a version
problem. The tester handles the encoding for you, but only if you told it the right
version.
Register: the live credentials handshake
status becomes registered. On failure it
records lastError, and that error is the thing worth reading.
Registration is where OCPI integrations get stuck, and the failures follow a pattern:
a side keeps presenting the dead registration token after the exchange, a partner never
fetches your endpoints, or the token encoding does not match the version. The
troubleshooting page walks each one.
Read your connections
404.
When their address changes
A partner who moves hosts, or an endpoint of your own behind a tunnel that hands out a new URL, does not need a new connection. Move the one you have:pending while we read their versions
endpoint at the new address, and returns to registered when that answers.
This is ours rather than OCPI’s. The spec’s “changing endpoints” procedure is a credentials PUT
that carries your versions URL, so it tells a partner that you moved. Nothing in OCPI tells a
client that the server moved, because a server’s versions URL is exchanged out of band in the
first place.