Infinitecampus api
Skill chrischall/infinitecampus-mcp/skills/infinitecampus-api
Query an Infinite Campus (Campus Parent portal) district directly with curl instead of running the infinitecampus-mcp server — log in with a real username/password, capture the session cookie + XSRF token, and curl grades, attendance, assignments, schedule, messages, documents, and fees. Use when you want IC data without the MCP, in a script, or on a machine where the MCP isn't installed. Infinite Campus is per-district: the base URL and district app name are configurable, not hardcoded.From its SKILL.md
npx -y skills add chrischall/infinitecampus-mcp --skill infinitecampus-apiAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.6 KB, as published. Nobody here has run it
Infinite Campus via curl (no MCP)
Infinite Campus's Campus Parent portal is a classic server-rendered site
(no public API, no bot wall). A plain form POST to verify.jsp logs a
parent in and hands back a session cookie — no browser bridge needed. This
is the same login infinitecampus-mcp's src/client.ts/src/auth.ts do;
this skill reproduces it with curl + a cookie jar.
Infinite Campus is per-district. Every parent's portal lives at their
own district's host (https://campus.<district>.k12.example.us or similar)
under an appName path segment. There is no shared IC endpoint — carry the
base URL and district from the same config the MCP uses:
: "${IC_BASE_URL:?set to your district's portal URL, e.g. https://campus.yourdistrict.org}"
: "${IC_DISTRICT:?set to the appName path segment, e.g. yourdistrict}"
: "${IC_USERNAME:?parent portal username}"
: "${IC_PASSWORD:?parent portal password}"
(Same 4 env vars the MCP reads — reuse the repo's .env or the host's
secret store; don't hardcode a district.)
One-time setup
None — curl and jq are the only tools needed (brew install jq if
missing). No profile/pairing step, no extension.
Login (capture cookies + XSRF token)
verify.jsp accepts a urlencoded form POST and returns 200 with an
<AUTHENTICATION>state</AUTHENTICATION> marker in the body — not
via HTTP status — plus ~20 Set-Cookie headers. Use a cookie jar so curl
does the Set-Cookie parsing/deduping for you:
JAR=$(mktemp)
BODY=$(curl -sS -c "$JAR" \
"$IC_BASE_URL/campus/verify.jsp?nonBrowser=true" \
--data-urlencode "username=$IC_USERNAME" \
--data-urlencode "password=$IC_PASSWORD" \
--data-urlencode "appName=$IC_DISTRICT" \
--data-urlencode "portalLoginPage=parents")
echo "$BODY" | grep -o '<AUTHENTICATION>[^<]*' # AUTHENTICATIONsuccess = good
password-error / account-locked / any non-success state means the
login failed — check that string before doing anything else. Credentials
go in the form body (never the query string — it lands in access logs).
Pull the XSRF token out of the jar (IC's JS reads it back and echoes it on
every request as X-XSRF-TOKEN; JSESSIONID alone is not enough):
XSRF=$(awk -F'\t' '$6=="XSRF-TOKEN"{print $7}' "$JAR")
Core call pattern
Every subsequent request rides the jar + the XSRF header:
ic_get() { # ic_get <path-with-query>
curl -sS -b "$JAR" -H "X-XSRF-TOKEN: $XSRF" -H 'Accept: application/json' \
"$IC_BASE_URL$1"
}
ic_get '/campus/api/portal/students' | jq .
A 401 means the session expired (IC sessions last ~5-6h) — re-run the
Login step to get a fresh $JAR/$XSRF and retry once. There is no retry
budget beyond that in this shell version (the MCP's CookieSessionManager
does this automatically; here just redo the login block).
The one rule: resolve student + enrollment first
Almost every read is scoped to a personID (student), and several need the
student's enrollment (enrollmentID / calendarID / structureID).
Always start with:
ic_get '/campus/api/portal/students' | jq '.[] | {personID, firstName, lastName, enrollments}'
Take personID for studentId/personID params, and the first (or
relevant) enrollments[] entry's enrollmentID/calendarID/structureID
for the endpoints that need them (attendance, school days, assessments,
features). There is no name→id search — ic_list_students (i.e. this call)
is the resolve step.
Endpoints
All 17 read endpoints (grades, attendance, assignments, schedule, messages,
documents, fees, teachers, assessments, feature flags) with ready-to-run
curl+jq recipes are in references/endpoints.md. Two (behavior,
food_service) have real paths that were never confirmed live against a
district with that module enabled — try them, expect a 404 if the module
is off or the district's shape differs.
Downloads
ic_download_document's url field (from the documents list) may be a
relative /campus/... path or an absolute URL. Same jar + header pattern,
just write to a file instead of piping to jq:
curl -sS -b "$JAR" -H "X-XSRF-TOKEN: $XSRF" -o report-card.pdf "$IC_BASE_URL$DOC_URL"
FeatureDisabled districts
Some districts turn whole modules off (behavior, food service, assessments,
documents, attendance). The MCP checks displayOptions first and falls
back to treating a 404 as "disabled." Doing the same by hand: fetch
/campus/api/portal/displayOptions/{structureID}?personID=X (see
references) and check the flag before assuming a 404 is a bug.
Notes
- Read-only in practice: every recipe here is a GET.
ic_get_messagefetching a message body may mark it read on some district configs — the MCP's own probe couldn't confirm the side effect either way. - No CUPS linked-district switching in this skill (a parent with kids in
2+ IC instances under shared SSO) — that's a multi-step token-exchange
flow (
src/client.ts'sdiscoverLinkedDistricts); out of scope here. Run the login flow again against the linked district's ownIC_BASE_URL/IC_DISTRICTif you need it. - This project is developed and maintained by AI (Claude).