Migration

anymd was called pdf-reader-mcp, then Citra. Existing setups keep working; this page lists what changed and how to move to the new names.

Packages

Before	Now
npx -y @sylphx/pdf-reader-mcp (bin pdf-reader-mcp)	npx -y @sylphx/anymd (bin anymd)
npx -y @sylphx/citra (bin citra)	npx -y @sylphx/anymd
MCP registry io.github.SylphxAI/pdf-reader-mcp, io.github.SylphxAI/citra	io.github.SylphxAI/anymd
@sylphx/citra-<platform> native packages	@sylphx/anymd-<platform>
CITRA_RUST_BIN	ANYMD_RUST_BIN (the old name is still read by the SDK)
SDK class Citra	Anymd (Citra stays as a deprecated alias)

@sylphx/pdf-reader-mcp and @sylphx/citra are published at the same version as thin aliases that run anymd, so old configs need no change. For new installs, and to get a shorter server name in your client, switch the command to npx -y @sylphx/anymd.

--allow-dir and MCP_PDF_ALLOWED_DIRS are unchanged.

Tools

Before	Now
read_pdf	read
search_pdf	search
pdf_evidence	inspect (same operations)
pdf_compare	inspect with operation: "compare"

The old names still work for this major version but no longer appear in tools/list, so agents discover the new ones.

The new tools take simpler arguments:

// before
{ "sources": [{ "path": "report.pdf", "pages": "1-5" }] }
// now
{ "source": "report.pdf", "pages": "1-5" }

search takes plain strings in sources (files, folders, or URLs) and searches every supported format, not just PDFs.

Markdown by default

read and search answer in compact Markdown: <!-- page N --> anchors, a small front-matter header, pipe tables, and a max_tokens budget (default 20000) with a cursor. On Attention Is All You Need, that is 42 KB instead of a 2 MB JSON envelope.

Structured JSON

If you relied on the JSON result (document map, elements with bounding boxes, tables with geometry, trust and accessibility reports), it is still there:

• Now: inspect with operation: "structure" and a profile of fast (default), quality, or research:

  { "operation": "structure", "sources": [{ "path": "report.pdf" }], "profile": "research" }

• Legacy read_pdf: passing profile (fast, quality, research) or any include_* flag still returns the JSON result. A plain read_pdf call with only sources now returns Markdown.

• Legacy search_pdf: pass detail: true for the JSON result with match geometry and provenance instead of the one-line-per-hit list.