PHI Greek Inscriptions
A search and text server for the Greek inscriptions of The Packard Humanities Institute. It is made for assistants such as Claude, ChatGPT and Grok, and for programs. This page tells a person what it does and what its answers mean.
What it holds
About 230,000 Greek inscriptions from 15 regions, as the printed editions give them: Inscriptiones Graecae (IG), the Supplementum Epigraphicum Graecum (SEG) and the regional corpora. Each text has its place, its date where the edition gives one, and its edition.
The texts are Greek only. There are no translations, no photographs and no lemma (dictionary word) search. An assistant that gives you English has made the translation itself.
How to connect an assistant
The address of the server is
https://epigraphy-mcp.packhum.org/mcp
In the settings of the assistant, add a connector (some assistants call it an MCP server, a tool or an app) and give it this address. No key and no sign-in are needed. In Claude: Settings, Connectors, Add custom connector. In Claude Code: claude mcp add --transport http phi-greek-inscriptions https://epigraphy-mcp.packhum.org/mcp
Then ask in plain words. The assistant decides when to call the server; to be sure, tell it to use PHI Greek Inscriptions.
What you can ask an assistant that is connected
- One inscription: “Show me IG I³ 10.”
- Where a word stands: “How many inscriptions have δῆμος, and in which regions?”
- Two things together: “Where does θεοῖς stand in the same line as ἀνέθηκεν?”
- By place, date and kind: “List the dated inscriptions from Delphi before 600 BC, oldest first.”
- What is on the stone: “Is this word read on the stone, or restored by the editor?”
Ask the assistant for the PHI number of each inscription it names (for example PH229506). That number never changes, and https://epigraphy.packhum.org/text/ followed by the number opens the text.
The search finds letters, not words or meanings
- It needs Greek. Nothing is translated. A search for “Augustus” or “gladiator” finds nothing: the assistant must give the Greek (Σεβαστός, μονομάχος).
- Latin letters are a code for Greek letters, not English spelling. The code is called betacode:
h is η, w is ω, q is θ, c is ξ, f is φ, x is χ, y is ψ. So dhmos is δημος, and demos is δεμος, a different and much rarer spelling. When a search is in Latin letters, the answer says how it was read and what the other readings give.
- It matches a string inside longer words. δημος also finds δημοσιον. An assistant can ask for a whole word, or for the start of a word to get all the endings of a stem.
- A stem also finds names. Διονυσ- finds the god Διόνυσος and, far more often, the man's name Διονύσιος. The answer lists each word that was found with its count: read that list before you trust a total.
- Spelling changes with place and date. Early Attic writes ε for ει and ο for ου; Doric has δᾶμος for δῆμος. One spelling does not find the others.
- Accents, breathings, capitals, brackets and line ends are ignored.
How to read the answers
- Brackets are the editor's.
[ ] encloses letters the editor restored, and a dot under a letter marks an uncertain reading. A search can be limited to the letters read on the stone, or to the restorations. A claim about what a stone says needs that distinction.
- Kinds of text are known for about a quarter of the corpus. “Epitaph”, “decree” and the like come from a short note by the editor, and the older IG volumes have none. A count by kind is a lower limit, never the full number. The answer says how much of the selection the filter could see.
- A date is often a range of years, and many inscriptions have no date. “The oldest” means the earliest start of a range; compare the ranges. Years BC are given as negative numbers to the server.
- Places are region, subregion and site. Athens is a site in the region Attica. A text is filed under the region of its publication, so a few Athenian dedications at Delphi are under Attica.
- Lines are the lines of the edition as the corpus divides them.
- A short context is not a quotation. A search shows a few words round each match. To quote, the assistant must open the inscription.
- Counts are complete up to 300,000 matches. Past that, the answer says “more than”, and the counts cover only what was reached.
- The order is the order of the corpus, by publication, unless a sort by date is asked for.
For programs
No key is needed. Each client address has a daily allowance of 2,000 inscriptions opened. Searches and lists do not use it. When the allowance is used, the server refuses more texts until the next day and says so.
| Address | Use |
|---|
/mcp | Model Context Protocol server (Streamable HTTP). Six tools: search_inscriptions, get_inscription, get_inscriptions, list_inscriptions, list_regions, list_publications. |
/api/v1/search, /api/v1/browse, /api/v1/inscriptions/{phi} | The same data as JSON |
/docs | Description of the JSON API |
A reader PHI knows can have a key, which has no allowance: write to the address below. Send the key as Authorization: Bearer <key> or X-Api-Key: <key>. A connector that cannot send a header signs in with the same key through OAuth. A wrong key is refused.
A parameter that a tool does not have is refused, and the error lists the names that exist.
Terms of use
The terms are those of the PHI website. If you use this server, yourself or through an assistant, you agree to use it only for personal study, and not to make copies except for your personal use under “Fair Use” principles of copyright law.
The server is for reading and searching. It is not for making a copy of the corpus.
Questions
Write to [email protected].
The Packard Humanities Institute