CV parsing API for AI agents
Read CVs your agent can trust.
Send a PDF or DOCX to POST https://api.verbatimcv.com/v1/parse and get versioned
JSON in which every candidate preference comes with the exact quote it was read from.
Nothing is invented, hidden prompt injection is removed, and anything your agent should double-check is
listed in agent_guidance.confirm_before_acting.
- 50 CVs a month free, no card
- PDF and DOCX, including scans
- CVs never stored
$ curl https://api.verbatimcv.com/v1/parse \
-H "Authorization: Bearer $VERBATIMCV_API_KEY" \
-F file=@cv.pdf
// part of the response
"work_arrangement": {
"value": ["remote"],
"source": "stated",
"confidence": 0.95,
"evidence": "Remote only"
},
"timezone": {
"value": "Europe/Lisbon",
"source": "inferred",
"evidence": "Lisboa, PT"
},
"agent_guidance": {
"confirm_before_acting": ["preferences.timezone"]
}
Live example
See it on a real CV
A fictional candidate's CV in Portuguese and English, and exactly what the API returned for it.
What the API returned (full JSON):
| Preference | Value | Source | Quote from the CV |
|---|---|---|---|
timezone | "Europe/Lisbon" | inferred 0.6 | “Porto, Portugal” |
working_hours | | absent | — |
work_arrangement | ["remote"] | stated 0.9 | “Remote only; contract or permanent.” |
engagement_type | ["contract", "full_time"] | stated 0.9 | “Procuro trabalho remoto, em regime de contrato ou a tempo inteiro.” |
availability | | absent | — |
compensation | | absent | — |
relocation | | absent | — |
work_authorization | | absent | — |
Dates parsed by code, never more precise than the CV:
- Engenheiro de Software Sénior, Pagamentos Atlântico SA (fictional):
fev. 2022→2022-02(month) - Software Engineer, Douro Cloud Lda (fictional):
set. 2018→2018-09(month)
Total experience: 99 months (sum of non-overlapping ranges; year-only dates counted from January). Confirm with a person before acting: preferences.timezone.
What you get
Facts an agent can act on, not a guess
Every value quoted from the CV
Each preference is stated, inferred or absent, with a confidence and the exact evidence quote. Missing facts are null, never guessed.
Prompt injection removed
White, tiny, off-page and invisible text, and instructions aimed at AI, are stripped before extraction and reported under safety.
Dates parsed by code
"fev. 2022", "Spring 2020" and "03/2021" become ISO dates with a precision. A year stays a year; ambiguous day and month order is flagged.
Hard layouts handled
Two-column and Canva CVs, text boxes, headers and footers, tables and scans (OCR), in many languages and scripts.
No protected characteristics
Age, date of birth, gender, photos, marital status and nationality are never returned; only their presence is noted.
Built for agents to act on
A versioned schema with fixed enums, and confirm_before_acting, generated by code, listing what to check with a person.
Use it from
Wherever your agent or workflow runs
Any HTTP client
curl, Python and JavaScript examples in the quickstart.
MCP
uvx verbatimcv-mcp gives Claude Desktop and other agents a parse_resume tool. In the official MCP Registry.
n8n
An importable Gmail → VerbatimCV → route workflow.
Apify
Parse lists of CV links into a dataset; connect it to Zapier or Make.
Try it on your own CVs
Free for 50 CVs a month, no card. Agents can start signup with POST /v1/signup. Paid plans from $29 a month (pricing).