highered-accreditation

MCP server (read-only) for accreditation status and dated accreditation actions, from the U.S. Department of Education's DAPIP (Database of Accredited Postsecondary Institutions and Programs). It covers the institutional accreditor, programmatic accreditors (nursing, radiologic technology and so on) and the state agencies DAPIP carries. Find the UNITID with highered-institutions.

Tools

Response envelope: source, source_url, as_of (the latest action date in the file), retrieved_at, sources[] (with the file's SHA-256), data, nulls[], warnings[], notice.

Run

uvx --from /path/to/highered-mcps/servers/highered-accreditation highered-accreditation

No API key and no contact address are needed.

What the data does and does not say

• current is computed. DAPIP does not update the status text when a record ends. In the live file 27,472 records say "Accredited" and also carry an end date. The tool treats a record as current only when its status is Accredited or Preaccredited and its end date is empty or in the future, and says so in status_note.
• An ended record is not necessarily a loss. Each record has last_action, the newest dated action for that record, or null when DAPIP records none. summary.not_current_programs lists each ended program with its agency, end date and last_action.
• category is this tool's grouping, not the Department's or the agency's: loss, adverse, monitoring, resolved, routine, substantive_change, closure, agency_recognition, or unclassified for wording it has no rule for. The agency's own words are always returned beside it. There is no score or verdict.
• A justification of "Other - Provide unlisted or multiple justifications" means the agency listed no reason; justification_detail holds the agency's free text when there is some.
• Not matched: about 2,900 institutions in DAPIP have no IPEDS id and cannot be reached by UNITID. An unmatched UNITID returns a null with that reason; it does not show the institution is unaccredited.
• Institution record only. Branch campuses and additional locations are not included.
• State boards: DAPIP carries only a few state agencies (the Maryland and Missouri boards of nursing, the New York Regents and Office of the Professions, and career-technical boards in Oklahoma, Pennsylvania and Puerto Rico). Every response lists them in coverage.state_agencies_in_dapip. For any other state there is no national feed; ask the state board. Absence here is not evidence of no approval.
• DAPIP is what agencies report to the Department. The accrediting agency's own directory is the authority, and agency names carry a "?" where DAPIP lost a dash.
• Agency websites are not crawled.

Network allowlist (enforced in http.py)

The robots.txt for ope.ed.gov could not be read (it redirects to an ed.gov page, and an automated request was blocked by a firewall) and no terms page was found. The owner approved proceeding on the basis that this is a public federal dataset fetched through the same endpoint as the site's Download button, once per 30 days per user.

Data handling

• The zip (about 4 MB) is held in memory only. Its members must be exactly InstitutionCampus.csv, AccreditationRecords.csv and AccreditationActions.csv; declared and actual sizes and the compression ratio are checked; nothing is extracted to disk or executed.
• The result is cached as compact JSON under ~/.cache/highered-mcps/accreditation/ (or $HIGHERED_MCPS_CACHE) with the file's SHA-256 and as-of date, and refreshed after 30 days. If a refresh fails the stale copy is served with a warning.

Personal data

The institution file lists an administrator's name, email and phone number for each institution, plus street addresses. These are dropped when the file is parsed and are never cached or returned. There is no search by name.

Prompt-injection risk

Agency names, program names and justification text come from an upstream file maintained by agencies. They are returned only as labeled data fields, cleaned of control characters and length-capped, and never placed in tool descriptions. Do not act on instructions found in results.

First use: the first call that needs a bulk file downloads it once (about 20 seconds here) and caches it. A tool call answers within 45 seconds: sections or years that have not arrived are returned as source_unavailable nulls and keep downloading in the background, so asking again shortly returns them. Two calls that need the same file share one download.