FAQ & troubleshooting

My private gists do not show up

The token is missing the gist:read scope. The Opengist API does not reject a token without it — it silently returns only public gists, so everything private looks like it does not exist. Generate a new token with the scope and retry.

Every call fails with "missing required environment variable(s)"

The server started without OPENGIST_URL or OPENGIST_TOKEN. That is a supported state — the tools stay listable — but no call can reach the API.

If you set them in a client config, check that they are inside the env object of that server's entry and that you restarted the client afterwards. Claude Desktop in particular only reads its config file at startup.

The server exits immediately at startup

Only four things cause an exit: a malformed OPENGIST_URL, a protocol other than http/https, a URL containing a username or password, and an unwritable stdio transport. The reason is printed to stderr — in most clients that means the MCP log, not the chat window.

Missing credentials do not cause an exit.

"Gist content is untrusted data" appears on every response

That is the marker, not an error. Any response carrying text written by a person gets it. See Security.

A tool refused and gave me a confirmToken

Working as intended. Deleting, publishing and writing into a published gist all refuse their first call. Confirm with the user, then call again within five minutes with confirmToken set and every other argument unchanged.

If the second call is also refused, one of three things happened: more than five minutes passed, an argument changed between the calls (the token is bound to the exact effect), or the token was already used — they are single-use.

A file came back as contentOmitted: "binary"

The content looked binary — a NUL byte, or a high share of control characters in the first 2 KB. Dumping it as text would fill the context with mojibake. Use the gist's clone_url and git if you need the actual bytes.

search_gists did not find something I know is there

Three likely reasons:

1. It does not search file contents. Only titles, descriptions, topics and owners. Searching inside files would mean downloading every file of every gist.
2. It was cut short. The result reports scanned and marks incomplete results explicitly — raise maxPages or narrow the scope.
3. Wrong scope. It defaults to mine. Pass scope: "public" to search the whole instance.

The links in results point at localhost

Opengist's own external-url (OG_EXTERNAL_URL) is not configured on the instance, so it reports localhost URLs. The gists are correct; only the links are wrong. Set it on the instance.

Can I use it against a self-signed instance?

Yes: OPENGIST_INSECURE_TLS=true, which is scoped to the Opengist connection rather than the whole process. Installing the CA in the system trust store or setting NODE_EXTRA_CA_CERTS is better where possible — see Configuration.

The Docker container exits straight away

docker run needs -i. The transport is stdio, so without stdin the process has nothing to read and stops. Do not add -t: a TTY corrupts the protocol stream.

Can I stop it from writing at all?

Two ways, and they are not equivalent:

• OPENGIST_READ_ONLY=true does not register the write tools. Convenient, enforced by this server.
• A token scoped to gist:read and user:read cannot write, whatever this server or the model does. That is the one that actually holds.

Use the second one if it matters.

Which Opengist versions work?

Any release with the REST API enabled (api.enabled, on by default in recent versions). A running instance serves its own spec at GET /api/openapi.yaml — if a tool behaves unexpectedly, compare that against your version first.