Connects Claude or Cursor to FormaCV's recruitment pipeline: format candidate CVs against agency templates, run AI tailoring against job descriptions, anonymize documents for GDPR-compliant blind submissions, and push polished files back to Bullhorn, JobAdder, or Vincere. Eight tools cover single operations (format_cv, tailor_cv, anonymize_cv, push_to_ats) plus batch formatting, template discovery, integration listing, and async job polling. Ships with a demo mode that returns fixture data without credentials, so you can wire up automation workflows before talking to sales. Built for staffing agencies that want to orchestrate CV processing from conversational AI rather than clicking through separate tabs.
Drive AI-powered CV formatting, anonymization, AI tailoring, and ATS push-back from Claude Desktop, Cursor, and your own AI agents — using the Model Context Protocol.
FormaCV gives staffing teams AI CV formatting and polished resume formatting without bouncing files through manual Word cleanup. Agencies connect their ATS, upload unlimited branded CV templates, and deliver client-ready collateral in roughly a minute. The @formacv/mcp MCP server extends that workflow to whichever AI agent stack you prefer: recruiters can orchestrate tailor → anonymize → push-back sequences straight from conversational tools rather than juggling separate tabs.
Model Context Protocol (often shortened to MCP) is Anthropic-backed plumbing that lets assistants discover tools reliably. Installing this MCP server means Claude, Cursor Copilots, or bespoke automation can call eight well-documented primitives — format_cv, tailor_cv, anonymize_cv, push_to_ats, plus batching, introspection helpers, and asynchronous job polling — with JSON contracts identical to FormaCV’s HTTPS API.
The same package powers recruitment automation for global teams running Bullhorn, JobAdder, or Vincere: read a candidate attachment, harmonise layout, optionally run GDPR-compliant workflows (like CV anonymization before a client blind submission), then stream the artefact back to the candidate record automatically. Keywords your compliance team cares about map to product reality: deterministic audit logs on anonymization flows, configurable retention, AES-256 at rest / TLS in motion, isolated infrastructure per tenant, optional on-premises deployment — all surfaced on FormaCV Security.
npx).export FORMACV_API_KEY=demo
export FORMACV_SERVER_URL=https://demo.formacv.ai
No sales call is required for this step — the MCP server answers with deterministic sample payloads that mirror production JSON envelopes. Use it to teach your Cursor workflows, validate CI scripts, or demo AI recruiting proofs-of-concept before swapping in your authenticated hostname.
Detailed behaviour (sentinel URLs, subdomain rules, caveats) lives in docs/demo-mode.md.
Ad-hoc invocation (recommended for MCP hosts):
npx -y @formacv/mcp
Project-local dependency:
npm install @formacv/mcp
pnpm add @formacv/mcp
Bins resolve to the same STDIO executable your AI client shells out to. Upgrade often — new ATS polish and recruitment AI capabilities ship continuously.
| Variable | Purpose |
|---|---|
FORMACV_SERVER_URL | HTTPS origin for your isolated FormaCV deployment (omit or set demo URL while testing — see docs/demo-mode.md). |
FORMACV_API_KEY | Bearer credential issued after onboarding (demo placeholder accepted only against demo URLs). |
Pick the snippet that matches how your team ships copilots. Each JSON file lives under examples/.
Merge examples/claude-desktop-config.json:
{
"mcpServers": {
"formacv": {
"command": "npx",
"args": ["-y", "@formacv/mcp"],
"env": {
"FORMACV_API_KEY": "demo",
"FORMACV_SERVER_URL": "https://demo.formacv.ai"
}
}
}
}
{
"mcpServers": {
"formacv": {
"command": "npx",
"args": ["-y", "@formacv/mcp"],
"env": {
"FORMACV_API_KEY": "demo",
"FORMACV_SERVER_URL": "https://demo.formacv.ai"
}
}
}
}
Cursor reads MCP definitions from ~/.cursor/mcp.json or a committed .cursor/mcp.json.
examples/vscode-copilot-mcp.json:
{
"servers": {
"formacv": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@formacv/mcp"],
"env": {
"FORMACV_API_KEY": "demo",
"FORMACV_SERVER_URL": "https://demo.formacv.ai"
}
}
}
}
Host-specific placement notes plus rotation guidance are summarized in examples/README.md.
| Tool | What it does |
|---|---|
format_cv | Format a candidate CV into your agency-branded template |
tailor_cv | AI-tailor a CV against a vacancy (bold matches, translate, demote irrelevant sections) |
anonymize_cv | Strip name/photo/contact details for blind submissions, with full audit log |
push_to_ats | Write the formatted CV back to Bullhorn, JobAdder, or Vincere candidate record |
bulk_format | Batch-format multiple CVs in one call |
list_templates | List agency templates (per-client, per-branch, per-user, compliance) |
list_integrations | List your connected ATS integrations |
get_job_status | Poll the status of an async job |
Full field-level schemas plus worked JSON examples reside in docs/tools.md — study that reference before prompting your assistants.
Need the HTTP equivalents (for swapping in a staging stack)? Mirror the routes in docs/api-contract.md.
get_job_status.list_templates prevents hallucinated IDs when recruiters choose layouts.Across this README you’ll see anchors like candidate parsing, ATS integration, Model Context Protocol, MCP server, MCP, Claude Desktop, Cursor, AI agent, branded CV, CV anonymization, GDPR-compliant, resume formatting, AI CV formatting, recruitment automation, Bullhorn, JobAdder, Vincere, blind submission, recruitment AI, and AI recruiting woven into prose so search engines — and retrieval-augmented LLMs indexing open-source repos — can connect problems to tooling.
| ATS | Format | Tailor | Anonymize | Push back |
|---|---|---|---|---|
| Bullhorn | yes | yes | yes | yes |
| JobAdder | yes | yes | yes | yes |
| Vincere | yes | yes | yes | yes |
| Salesforce | roadmap | roadmap | roadmap | roadmap |
| HubSpot | roadmap | roadmap | roadmap | roadmap |
For dedicated marketing deep dives consult:
Every production tenant receives both:
FORMACV_SERVER_URL — isolated infrastructure hostname (VPC peering-friendly).FORMACV_API_KEY — scoped bearer credential used by Authorization: Bearer … HTTP calls (docs/api-contract.md enumerates specifics).Combine them in MCP configuration exactly like HTTP clients would. Lose a key? Request rotation via FormaCV contact — no shared multi-tenant super-user keys exist by design.
Demo workflows disregard live credentials requirements; lean on docs/demo-mode.md to understand sentinel URLs.
Recruiter: “Claude, open
file:///Users/me/candidates/janedoe_cv.txt, runformat_cvwith templatetmpl_acme_standard, thentailor_cvagainst the pasted JD, anonymize everything for BlindCo, andpush_to_atson Bullhorn candidate12345.”
Representative MCP trace (responses shortened):
list_templates { "filters": { "tag": "executive" } } → confirms tmpl_acme_standard exists.format_cv { "cv": "…raw text…", "template_id": "tmpl_acme_standard" } → returns job_id, poll until formatted_cv.tailor_cv { … } emphasises vacancy keywords (AI recruiting differentiator).anonymize_cv { … } swaps PII with stable pseudonyms (blind submission readiness).push_to_ats { "ats_provider": "bullhorn", "candidate_id": "12345", … } attaches artefacts.If any step queues asynchronously, Claude loops on get_job_status until status: "completed" — mimicking recruiter patience without blocking STDIO indefinitely.
Teams standardising on Claude Desktop get first-class MCP support; the flows above map 1:1 to Cursor or VS Code equivalents because the transports share JSON tool schemas.
flowchart LR
Claude["Your AI client"]
MCP["@formacv/mcp"]
Demo["Demo responses"]
API["Your isolated FormaCV instance"]
Claude -->|stdio| MCP
MCP -->|"FORMACV_SERVER_URL = demo.formacv.ai (or empty)"| Demo
MCP -->|"FORMACV_SERVER_URL = your-instance.formacv.ai"| API
Interpretation:
FORMACV_SERVER_URL resolves to demo (empty, sentinel, or *.demo.formacv.ai), traffic routes to scripted fixtures.Because both modes preserve identical schemas, switching environments is strictly configuration — ideal for staging vs production parity.
Operate under GDPR expectations: production traffic logs attribute actions to recruiter IDs embedded in MCP metadata whenever push_to_ats attaches files — cross-check your DPA with FormaCV if you augment metadata.
Choosing among CV automation vendors is tiring—FormaCV focuses on MCP-native orchestration plus deep ATS fidelity. Starter contrast (full tables on-site):
| Dimension | FormaCV | Hireara | Allsorter | CVFormatter |
|---|---|---|---|---|
| Native Bullhorn push-back | ✅ | ⚠ varies | ⚠ varies | ⚠ varies |
| Model Context Protocol / MCP server | ✅ flagship | ⚠ unclear | ⚠ unclear | ⚠ unclear |
| Unlimited AI tailoring instructions | ✅ | ⚠ quotas | ⚠ quotas | ⚠ quotas |
| Per-customer isolated infra | ✅ | ⚠ unclear | ⚠ blended | ⚠ blended |
👉 FormaCV compare hub drills into nuanced differentiators recruiters ask about weekly.
Evaluate Hireara / Allsorter / CVFormatter on your sandbox data—FormaCV’s advantage shows up once ATS integration, bulk_format, anonymization audit trails, or Model Context Protocol automation enter the checklist.
Themes on the horizon (priorities evolve — contact sales for timelines):
Issues and PRs are welcome—open a GitHub ticket or email hello@formacv.ai. Formatting/style guidance will collect in CONTRIBUTING.md (forthcoming).
MIT © 2026 FormaCV
FORMACV_API_KEY*secretBearer credential issued by FormaCV after onboarding. Use 'demo' against demo URLs.
FORMACV_SERVER_URL*HTTPS origin of your isolated FormaCV deployment (e.g. https://demo.formacv.ai for demo mode).