CIA CREST Database: How to Search the CIA Reading Room

The CIA CREST database is the largest public archive of declassified Agency records, and it is searchable free online. Most visitors type one word into the CIA Reading Room, scroll a few pages and leave. As a result, they miss most of what the tool can do. This guide explains what CREST is and how the search engine reads queries. It then shows how to decode CIA-RDP document numbers. Finally, it explains how to cite a PDF so others can find it again.

Every worked example below uses a real document number. Each one was checked against CIA-hosted records or archived copies of CIA pages. Where the live site could not be tested, the article says so plainly.

What the CIA CREST Database Actually Is

CREST stands for CIA Records Search Tool. It began as a full-text system for records reviewed under the Agency’s 25-year declassification programme. The CREST collection page ties that programme to Executive Order 13526, formerly EO 12958. The original order dates from April 1995. It requires the declassification of nonexempt, historically valuable records 25 years or older.

So CREST is not every CIA release. Instead, it is the electronic subset of records that passed through that review. Manually reviewed paper records went to the National Archives in original form. Beyond CREST, the Reading Room holds FOIA releases and curated historical collections that sit outside CREST.

CIA divides CREST into named collections. Its collection page lists Berlin Tunnel, Consolidated Translations, General CIA Records, Ground Photo Caption Cards, JPRS, Library of Congress, NIS, NGA Records (formerly NIMA), OSS Collection, Scientific Abstracts, Secret Writing and STARGATE. General CIA Records is by far the largest. A 2017 MuckRock count put it at 768,793 documents.

The content is broad. For example, CIA’s own launch notice lists finished intelligence from the 1940s to the 1990s. It also names Directorate of Operations reports, science and technology files, National Intelligence Council estimates, OSS records and imagery reports. Foreign translations and scientific abstracts also fill large parts of the database.

From Four Terminals to the Open Web

For most of its life, CREST was hard to reach. According to the Agency, it installed the system at NARA II in College Park, Maryland, in 2000. Its 2017 announcement says releases to CREST began in 1999. Either way, researchers had to travel to use it.

Access was narrow even on site. The Guardian and the BBC reported that just four computers held the collection. According to the BBC, the terminals ran only from 09:00 to 16:30. Even so, the CIA’s collection page notes that researchers printed about 1.1 million pages from the system over the years.

Pressure built outside the Agency. The Verge reported that MuckRock filed a FOIA case over CREST in 2014. In response, CIA had estimated that copying the database would take years and cost about $108,000.

That changed on 17 January 2017. On that day, the CIA announced that approximately 930,000 documents, totalling more than 12 million pages, were now in its Electronic Reading Room. It added that documents are regularly added as 25-year reviews continue. Record pages still carry this history in their metadata. MKULTRA Subproject 5, for example, shows a creation date of 15 December 2016, weeks before the public launch.

How the Reading Room Search Behaves

The public search lives at cia.gov/readingroom/search/site. A simple query becomes part of the address. For instance, a search for STARGATE resolves to /readingroom/search/site/STARGATE. Because of that, any result list is easy to bookmark and share.

An archived results page from January 2025 shows the working interface. A count sits at the top; that STARGATE query reported 12,488 items. A sort control offers Relevancy or Title, in ascending or descending order. The left column holds three filters: collection, release date and post date. Each filter shows a count in brackets. Clicking one then adds a parameter such as f[0]=im_field_collection to the URL.

Each hit also displays a compact metadata line. It shows document type, collection, document number, release decision, original classification, page count and dates. An OCR text snippet follows. After that, opening a hit leads to a record page with the full metadata and an attached PDF.

Insider Release attempted to run fresh test queries on 28 September 2026. The CIA site’s bot protection blocked automated access to search and document pages. For that reason, the interface description here relies on that archived page, the CIA’s own search guide and MuckRock’s 2017 guide. Readers using a normal browser should see the same layout, although CIA can change it at any time.

Search Operators That Change Results

The CIA’s Advanced Search Guide sets out the core syntax. The engine is not case sensitive for ordinary words. AND requires both terms. OR accepts either term, while NOT excludes the second term. The guide’s own example shows the scale: Stalin AND Hitler returned 1,643 documents, while Stalin OR Hitler returned 22,304.

Capitalisation matters for the operators themselves. MuckRock’s 2017 guide found that lowercase “and” is treated as a plain search word. It also warned that several unquoted words may return documents containing any one of them. Quotation marks fix that because they force an exact phrase. Parentheses group terms, as in (Nixon OR Ford) AND Kissinger.

Proximity searches are the most useful precision tool. Put a tilde and a number after a quoted phrase. The CIA guide’s example is “Stalin Hitler”~5, which finds the two names within five words. In addition, MuckRock documented a caret for weighting, such as Soviet^2 Union.

Wildcards help with spelling variants and OCR damage. An asterisk stands for zero or more characters. A question mark stands for one character. MuckRock notes that a wildcard cannot open a search term.

OCR is the hidden constraint. The CIA guide warns that scan quality limits full-text accuracy. Stained, faded or typewritten pages may not match a correct query. In practice, short and distinctive terms usually work better than long phrases.

Advanced Search Fields and Filters

The Advanced Search page combines full text with metadata filters. The CIA guide lists eight fields: Search Term(s), Collection(s), Document Title, Document Number/ESDN, Original Classification, Publication Date, Content Type and Case Number.

Document Title searches only the title field. That is useful because CREST titles are indexer summaries, often in capitals. Publication Date takes the format YYYY-MM-DD. It also supports conditions such as “Is between” for ranges.

Original Classification uses single letters. T means Top Secret, S Secret, C Confidential, R Restricted, U Unclassified and K Unknown. Content Type narrows results to forms such as cable, letter or memo. MuckRock’s review of the metadata found 93 content types, far more than the guide lists. Case Number, by contrast, applies only to FOIA releases.

Collection filters carry one trap. The CIA guide warns that some documents do not fall into any collection. A search limited to collections may miss them. MuckRock suggested excluding Scientific Abstracts and Consolidated Translations to cut false hits. Still, it flagged the same risk.

Reading CIA-RDP Document Numbers

Most CREST records carry an ESDN, labelled on record pages as “Document Number (FOIA) / ESDN (CREST).” The CIA guide explains that the number combines an archive job, box and folder with a document number assigned at review. Those location codes once tracked the paper record in CIA storage.

The guide’s sample is CIA-RDP54-00252A000100010015-2. CIA is the agency. RDP is the prefix. 54-00252A is the job number, and its first two digits mark the year the records were retired. The next groups are box 0001, folder 0001 and document 0015-2.

That structure enables folder-level searching. To use it, the guide shows wildcard patterns for each level. *33-02415A* finds everything in job 33-02415A. Next, *33-02415A0001* narrows to box 1. Finally, *33-02415A00010001* narrows to folder 1 of that box.

Not every record has an ESDN. The guide says documents not retrieved from the archive lack one. FOIA and special-collection releases often use numeric IDs such as 00746209. Other patterns include DOC_ numbers and LOC-HAK identifiers for Henry Kissinger’s Library of Congress papers, such as LOC-HAK-464-6-1-9. Each one still works as an exact search key.

Worked Examples: UFOs, Stargate and MKUltra

Start with UFOs. The best-known CREST UFO record is the 1953 Robertson Panel material. The CIA-hosted report of the Scientific Advisory Panel meeting of 14–18 January 1953 carries ESDN CIA-RDP81R00560R000100030027-0. Under the CIA’s wildcard rules, entering the job stem *81R00560R* in the document-number field should return other records from the same retired job. That is a quick way to find the surrounding paperwork.

Stargate, by contrast, shows the value of collection filters. A search for STARGATE, narrowed to the STARGATE collection, returned 12,471 records in the January 2025 snapshot. CIA-RDP96-00789R003000110001-6, “A Program Plan for Project Star Gate,” is a typical hit. Insider Release’s guide to the CIA Stargate documents maps that collection in depth.

MKUltra shows why ID formats matter. CIA-RDP78-04913A000100030026-4 is a two-page memo titled “Project MKULTRA, Subproject 5,” dated 28 May 1953. Its record page lists document type CREST, collection General CIA Records, original classification S and a release date of 18 September 2002. The ID decodes as job 78-04913A, box 0001, folder 0003, document 0026-4.

A related file, MKULTRA Subproject 7, uses a different key. Its record, 00746209, is typed SPECIALCOLLECTION and runs to 33 pages. So a search limited to CIA-RDP strings would miss it. Running both the keyword and the numeric ID family catches both.

How to Download and Cite a CREST PDF

Every record page links an attached PDF. The file usually shares the record’s ID. Its address follows the pattern cia.gov/readingroom/docs/ plus the ID and .pdf. Meanwhile, the record page uses a lowercase version under /readingroom/document/. A reliable citation routine takes four steps.

1. Copy the full document number exactly, including the final digit after the hyphen.
2. Record the title, publication date and original classification from the record page.
3. Save both the record-page URL and the direct PDF URL.
4. Download the PDF and keep a local copy, since links and access rules can change.

A clean citation then reads like this: “Project MKULTRA, Subproject 5,” 28 May 1953, CIA-RDP78-04913A000100030026-4, CIA FOIA Electronic Reading Room. The ESDN matters more than the URL, since it also appears on the scanned page itself, in the “Approved For Release” stamp.

When the Search Fails

The Reading Room sits behind bot protection. During this test, scripted requests were redirected or denied. A plain browser session, without scripts or aggressive privacy extensions, is the first thing to try.

Archived copies are the second fix. The Internet Archive holds many Reading Room pages and PDFs. They help confirm that a document existed at a given address. Even so, the CIA copy should be the cited source whenever it loads.

Zero results do not prove absence. OCR errors, indexer spelling and missing collection tags can all hide a record. So try synonyms, period jargon, wildcards and title-only searches before concluding anything. MuckRock’s example is apt: officials wrote “counsel” far more often than “lawyer.”

Finally, some records are simply elsewhere. Manually reviewed CIA paper sits at the National Archives. Newer statutory collections have their own tracks, such as the NARA UAP records collection in Record Group 615. A FOIA request remains the route for material never released.

INSIGHT: The CIA CREST database rewards researchers who search by structure rather than by headline. Exact phrases, proximity operators and ESDN wildcards turn a noisy keyword list into a folder-level map. The most durable research habit is citing the document number, not the screenshot.

FAQs

The questions below cover the most common CREST search problems.

What is the CIA CREST database?

CREST, the CIA Records Search Tool, is a full-text archive of CIA records declassified under the 25-year review programme. It was once available only on terminals at the National Archives in College Park, Maryland. Since January 2017, it has been searchable in the CIA’s online Electronic Reading Room.

How many documents are in CREST?

At the January 2017 launch, CIA said about 930,000 documents totalling more than 12 million pages went online. The Agency says new records are added as reviews continue. Its collection page separately cites more than 11 million pages released in electronic form.

What does CIA-RDP mean in a document number?

It is the start of an ESDN. CIA marks the agency and RDP is the prefix. The rest encodes the archive job, box, folder and document number that tracked the paper original.

Which search operators work in the CIA Reading Room?

The CIA’s guide documents AND, OR and NOT, tilde proximity searches and wildcards. MuckRock’s guide adds quotation marks for phrases, parentheses for grouping and a caret for weighting. Operators should be typed in capitals.

Why does a CREST search miss documents I know exist?

OCR errors, unusual title spelling and missing collection tags are the usual causes. Search by exact document number, use wildcards, and remove collection filters before assuming a record is gone.

Takeaways

CREST is the 25-year declassification archive that the CIA moved online in January 2017. It launched with roughly 930,000 documents and more than 12 million pages. The Reading Room search supports Boolean operators, phrases, proximity and wildcards. Filters cover collection, release date and post date.

The CIA-RDP number is the researcher’s strongest tool. It identifies a document exactly. It also reveals its original job, box and folder. Wildcards on those segments open whole folders. Other ID families, however, such as numeric FOIA IDs and LOC-HAK numbers, need separate searches.

Good practice stays simple. Search precisely, cite the document number and keep a local copy of every PDF.

Call to Action

If you have tested a CREST search technique that works better than those described here, share the query and a Reading Room link. Prefer cia.gov record pages over mirrors when both exist. Insider Release will keep updating its declassified-document finding aids as the Reading Room changes. Corrections grounded in CIA PDFs are welcome. Speculation without a citation is not.

Disclaimer: This article was created with the partial or full assistance of artificial intelligence. The text and all accompanying images were generated or significantly supported by AI tools.

Insider Release — Declassified. Analyzed. Explained.