Give your agent a CV parsing tool
Install the VerbatimCV MCP server and your agent gets a parse_resume
tool. It takes a local file path or a public URL and returns JSON in a fixed schema, where every
preference comes with the exact quote from the CV and anything uncertain is listed in
agent_guidance.confirm_before_acting.
Why a tool beats reading the PDF
- Layout: agents see PDF text in extraction order, which scrambles multi-column CVs. The tool fixes reading order first.
- Hallucination: the tool returns
nullwithsource: "absent"when the CV doesn't say something, instead of a plausible guess. - Prompt injection: CVs increasingly contain hidden text aimed at AI screeners. The tool strips it before extraction and reports it under
safety, so it never reaches your agent as instructions. - Consistency: the same CV gives the same fields and enums every time, so the agent's next step doesn't depend on how it phrased a prompt.
Claude Desktop setup
Add this inside "mcpServers" (Settings → Developer → Edit Config) and restart:
"verbatimcv": {
"command": "uvx",
"args": ["verbatimcv-mcp"],
"env": { "VERBATIMCV_API_KEY": "vcv_your_key_here" }
}
Other MCP clients use the same command and environment variable. The tool only reads PDF or DOCX files up to 10 MB, so an agent can't be tricked into uploading other files.
If your agent sets things up itself
An agent can request a free key for its user with POST https://api.verbatimcv.com/v1/signup
and {"email": "..."}. The user opens the emailed link to see the key.
Try it
Get a free API key (50 parses a month, no card), then:
curl https://api.verbatimcv.com/v1/parse \
-H "Authorization: Bearer $VERBATIMCV_API_KEY" \
-F file=@cv.pdf
Agents can use the parse_resume tool instead: uvx verbatimcv-mcp. See the
quickstart for Python, JavaScript and MCP setup.