One line in a system prompt — validate the DOI before citing, drop the citation if validation fails — turns a fabricated reference from something a reader catches into something the model catches. The credential design deserves a note too: the setup command can put the key in the OS credential store and leave only a reference in the client config, and it refuses to overwrite an existing credential unless you explicitly ask. Batch tools exist for a reason; a bibliography checked one call at a time will burn a day's quota fast.
A citation layer for models that invent references. It checks whether a DOI actually exists, resolves messy citation text back to the real paper, formats references in the style a journal wants, and keeps saved collections you can import into and export out of. It returns citation metadata only — not PDFs or full text.
- A DOI checked for existence, a book looked up by ISBN, a paper recovered from garbled citation text, and up to 50 citation strings resolved in one request — `validate_doi`, `lookup_isbn`, `reverse_lookup`, `batch_resolve`
- Corpus search with author, category and citation facets, plus an API health check — `enhanced_search`, `health_check`
- A DOI formatted in any of 2900+ CSL styles, several formatted at once, a list of DOIs batch-checked, styles found by name or paged through, and grouped in-text markers generated — `format_citation`, `batch_format`, `verify_references`, `search_styles`, `list_styles`, `group_cite`
- ORCID profiles searched and fetched, and a profile's publications indexed so they become searchable — `orcid_search`, `orcid_profile`, `ingest_orcid`
- Signed-in collections: list, add one or many, import BibTeX and RIS, export as BibTeX, search within, check duplicates, correct an entry's metadata, merge entries, tag, reorder, remove and delete — `list_collections`, `add_to_collection`, `batch_add_to_collection`, `import_bibliography`, `export_collection`, `search_collection`, `check_duplicates`, `update_entry_metadata`, `merge_entries`, `update_tags`, `reorder_collection`, `remove_from_collection`, `update_collection`, `delete_collection`
- Sharing and cross-collection moves — a shareable link created, revoked or viewed by token, collections merged, entries moved — `share_collection`, `unshare_collection`, `view_shared`, `merge_collections`, `batch_move_entries`
- Your plan and remaining lookups — `usage`
No key for basic lookup and formatting: an anonymous server gets 20 lookups/day, and a free account raises that to 60. Collection tools need `OOKCITE_API_KEY`. One command sets everything up — `npx @turtletech/ookcite-mcp setup` detects supported clients and writes their config, and `setup --connect` creates the Free plan and stores the key in the platform credential store so only a reference goes into client config. The package is `@turtletech/ookcite-mcp` (0.6.1) on npm, also installable through `cargo binstall ookcite-mcp` or as a prebuilt binary. Key lookup follows `OOKCITE_API_KEY`, then `OOKCITE_API_KEY_COMMAND`, then `OOKCITE_API_KEY_FILE`, then the platform credential; with none set the server starts anonymous. Set `OOKCITE_MCP_READ_ONLY` to `1` to hard-disable every collection mutation.
One command plus a key — npx @turtletech/ookcite-mcp setup, then supply credentials