[P1] Remove pdf from the input_format enum #47

Links

• Epic: Tracking: gaps surfaced by an agent-driven document workflow (from #40) #44
• Evidence: docs/audits/2026-08-agent-workflow-audit.md → Findings 1 and 3
• Code: src/mcp_pandoc/server.py L114

Problem

input_format advertises pdf. Pandoc has no PDF reader, so this can never succeed.

$ pandoc -f pdf file.pdf -t markdown
Unknown input format pdf
Pandoc can convert to PDF, but not from PDF.

Checked the last 25 pandoc releases; none added a PDF reader. Unlike the pptx case, this is a general statement rather than a property of one pandoc version.

Why it matters

The input schema is a promise made to a language model. A model reading pdf in the enum will attempt PDF input, fail, and often retry the same call.

Severity is P1 rather than P0: calls fail cleanly, with no data corruption and no startup impact.

Acceptance criteria

• pdf removed from the input_format enum
• If requested anyway, return an isError result naming the limitation rather than passing it to pandoc
• Test asserting the enum contents
• README distinguishes native readers from native writers

Related

• Reading PDFs at all would need a non-pandoc dependency. Tracked separately as a scope decision.
• The same directional problem affects txt, tracked separately.