Integrations · Model Context Protocol
Preflight documents from any AI agent
The Preflight MCP server lets Claude, Cursor, Cline, or any MCP-capable agent preflight, validate, and check your production documents — PDF, AFP/MO:DCA, PCL, PostScript, IPDS — plus decode USPS Intelligent Mail barcodes and compute effective print DPI, as callable tools.
Local (stdio)
Run the server with npx and point your agent at it. The two offline tools work with no key; the analysis tools need a Preflight API key (create one in your workspace).
Hosted (Streamable HTTP)
Prefer not to install anything? Point an MCP client at the hosted endpoint and authenticate with your API key:
Tools
| Tool | Key? | What it does |
|---|---|---|
| preflight_analyze_document | API key | Preflight a PDF/AFP/PCL/PostScript/IPDS file — findings + per-check coverage. |
| preflight_get_result | API key | Fetch/poll an analysis result by id. |
| preflight_list_checks | API key | List recent checks for your workspace. |
| preflight_list_check_types | no key | What Preflight checks, and which formats it parses. Offline. |
| preflight_decode_imb | no key | Decode + validate a USPS Intelligent Mail barcode. Offline. |
| preflight_calc_effective_dpi | no key | Effective print DPI (pixels ÷ printed size). Offline. |
What it will not pretend to know
Every result carries the same honesty contract as the rest of Preflight: anything a check cannot determine from the file is returned as “not evaluable” with a reason — never folded into a pass. If a document can’t be parsed, the tool says so; it does not invent findings. The API key and your file bytes are never logged. Preflight validates and reports — it does not convert, repair, or certify documents.