Connect your Claude client to your Jarvi pipeline over MCP. Search, draft and act on your candidates from claude.ai, Claude Desktop or Claude Code, within your scopes, with everything logged.
The gap the integration closes.
Generic. You still dig the bios, pull each resume, then rewrite the email yourself.
Hi Marc, sharing five strong Senior React profiles for the Lead Frontend role. Headline: Camille D. (7y React, available in 15 days). Resume and full shortlist attached, happy to set up first calls this week.
Grounded on your live pipeline. Drafted in seconds. Sent through Jarvi in one click.
One chat window, your real data. Search, create, fill fields, score, message and report. Real actions, written back to Jarvi and audited.
Claude is not limited to chatting or messaging. It creates jobs, companies and contacts in Jarvi, fills your custom fields, links a company to a project, moves a candidate to the next stage and chains several actions in a row. Hand it a LinkedIn job post and it sets up the opening, attaches the company, adds the right contact and prepares the shortlist, every step written back to your ATS and logged.
Tell Claude you need five senior React engineers in Lyon, open to a move, with a fintech background. Back come real people from your Jarvi pool, ranked by fit, each with its last touchpoint shown. There is no boolean syntax to remember: your plain-English brief runs against the very index your recruiters already search inside Jarvi. Refine it out loud, drop anyone contacted in the last six months, and the list updates instantly.
Point Claude at a candidate and it opens the fiche, reads what matters (current role, how they replied before, where they sit in your projects) and writes a first message that sounds like you. Approve it and it goes out through the Jarvi channel you already use (LinkedIn, email, WhatsApp, Telegram, SMS), logged on the candidate timeline. Nothing lives in a stray chat tab; the thread stays on the fiche.
Hand a project ten applicants and ask for a ranking against the brief. Claude reads the project from Jarvi, opens each fiche, and returns an ordered shortlist with reasons tied to the criteria you set, not generic fluff. Keep asking follow-ups without re-pasting anything. When you are happy, the scores and the rationale write back to Jarvi for the rest of the team.
Ask for the weekly report on the Globex search and get a structured update with active candidates, recent moves, blockers and next actions, generated from the live project data. Paste an interview transcript and Claude returns a structured debrief, moves the candidate to the right stage, notifies the assigned recruiter. Same logic for client P&L summaries, pipeline rollups, or your own weekly review.
Watch one request turn into real work: Claude creates records, enriches them and carries the search forward, step by step.
Spotted a role on LinkedIn? Drop the post into Claude and tell it to take the whole thing from there.
Claude opens the job in your ATS and fills each field from what the post says.
It tracks down the hiring company, finds its page and attaches it to the project in Jarvi.
It works out who actually published the role on LinkedIn, the person worth reaching.
It infers the company email convention from their site (think firstname.lastname@) and composes the address.
A contact is added under that company in your ATS, email included.
It pulls the candidates from your pool who match the role best and ranks them.
Each shortlisted profile gets its own polished report.
It tailors an approach for every profile and emails each report, angled per person.
A typical multi-step run, and the point of MCP: Claude structures records, fills fields and chains actions across your ATS, well beyond sending a message.
A short map for buyers comparing approaches.
In a 30-minute demo we plug Claude into your own Jarvi workspace and let it search, draft and act on your live data over MCP.
One open standard. No middleware. No proprietary lock-in.
claude.ai, Claude Desktop or Claude Code. Any MCP-compatible client.
Hosted by Jarvi. OAuth 2.1 with PKCE.
Your candidates, projects, channels and audit log.
No engineering, no middleware. Open Claude, add the MCP URL, go.
From Settings, copy the MCP endpoint of your workspace. Permissions follow your existing Jarvi roles.
Open claude.ai, Claude Desktop or Claude Code, paste the URL as a connector, sign in to Jarvi via OAuth 2.1 with PKCE. One screen.
Type "Find my five strongest Senior React profiles in Lyon" and watch Claude pull your live data, then draft the next action.
Need the full setup guide? Read the docs.
Copy, paste into Claude with the Jarvi MCP connected. Done.
Rank my top 5 candidates for the Senior React role in Lyon and explain each match.
Score the ten applicants on the Senior PM project against the brief, with one line of rationale each.
Write a LinkedIn opener for this profile in my voice, referencing her current role.
Nudge everyone on the Globex search who has stayed silent for a week.
Find lookalikes of the candidate I placed at Initech last quarter and propose a first message.
Where is Project Phoenix at? Volume, response rate, time-to-fill, blockers, next action.
Build the weekly progress report for ACME, ready to send by email.
Turn this five-line brief into a polished job description and push it to Jarvi.
Prepare three interview questions tailored to this candidate fiche and the role brief.
Pull ten near-misses worth re-approaching on the Senior Data role.
Spot duplicate profiles in my Acme pipeline and merge each onto the richest fiche.
Write a kind rejection email for the five candidates we passed on for Acme.
Claude only sees what your Jarvi permissions allow.
There is no static API key sitting around to leak. Each person authenticates through OAuth, inheriting the Jarvi permissions they already have, and you can revoke any session from settings.
Data is stored on certified European infrastructure, encrypted in transit and at rest, GDPR compliant. SSO available on Google or Microsoft enterprise mail. Each customer's data is isolated.
Every exchange with Claude is recorded: who asked, what was read, what was written, on which channel and when. The log lives in Jarvi and exports on request.
The MCP server respects the Jarvi roles you already configured. Someone only reaches the candidates and projects they could already open in the app.
2026-05-07 09:14:21 user=marie@agency.fr
prompt : "find top 5 React profiles Lyon"
read : candidates(7), projects(1)
scope : own_projects
action : none
2026-05-07 09:14:43 user=marie@agency.fr
prompt : "draft submittal for ACME"
read : candidates(3), client(ACME)
scope : own_projects
action : email_draft.created (id: dft_8a2)
2026-05-07 09:18:02 user=marie@agency.fr
approve : email_draft.dft_8a2
action : email.sent (id: msg_193)The audit trail captures every Claude exchange: who asked, when, through which channel, which fields were read, which action was executed. Modifications are tracked too.
Searchable log, exportable as CSV, retained for the full lifecycle of your workspace.
"Why AI works 1000x better than mail merge"
You know the drill… You think: “I’ll just send my emails oldschool via Outlook + Word + Excel.” Because yes, that has been working for 20 years.
Well… I proved again why I NEVER do that anymore. Mail sent without a signature? Without a first name? One person even got the same wrong email 3 times?
And then anyone dares to say that AI is “not reliable”? Don’t make me laugh. AI sends my messages perfectly personalized, without errors, without hassle. Jarvi has never forgotten a single first name or sent the same email to someone three times.
I’d rather work smart than retro.
"2 of us do what others need 5 or 6 for"
“How many people do you have to manage all this?” That’s the question a prospect asked me a few days ago for a recruitment mission. My answer: there are two of us. And I’m proud of it.
Why? Because with two people, we accomplish what others sometimes need five or six for. Not by magic. Not by burning out. But because everything is thought through, structured, and properly equipped. With an ATS that also serves as a CRM, allowing us to track every candidate, every step, every company.
Yes. The Claude ATS integration speaks the Model Context Protocol, Anthropic's open standard for connecting LLMs to live data. Jarvi hosts and maintains the MCP endpoint on your behalf, no devops, no plugin to install.
Any MCP-compatible client: claude.ai (web), Claude Desktop (macOS/Windows), Claude Code (CLI), and any other client that supports MCP connectors. Same Jarvi endpoint, same scopes.
Not to use AI inside Jarvi. The Jarvi AI is built into the product, powered by Claude Sonnet 4.7, and included in every Jarvi plan. To use Claude externally over MCP, you connect with your own Claude account.
Claude Sonnet 4.7. When you use the Claude ATS integration with your own Claude client, you pick the model on your side (Sonnet for daily work, Opus for deep analysis).
When you use the AI inside Jarvi, usage is included in your Jarvi plan, no markup, no per-prompt fee. When you connect your own Claude client over MCP, your Anthropic consumption is billed by Anthropic directly.
On infrastructure based in London. Data is encrypted in transit and at rest, on SOC 2 Type 2 certified servers, GDPR-compliant. Each client's data is isolated, and Jarvi never uses it for anything else.
When used through Jarvi: no. Jarvi has a contractual agreement with Anthropic, no storage of conversation data, no training, EU hosting. When you connect your own Claude client, the answer depends on your Anthropic plan and configuration.
Yes. Access scopes are configurable per user and per project. Hiring managers and collaborators only see the projects they are invited to. Larger organizations can request tailored access rules.
MCP access is revoked instantly. Audit log entries are kept for traceability. No further data is sent to Anthropic after disconnection.
Yes for ChatGPT: MCP is gaining adoption beyond Anthropic, and compatibility with non-Anthropic MCP clients is rolling out as they add native support. Make and Zapier are supported as standalone automation channels around Jarvi.
They share the same Claude model under the hood and write to the same audit log. The Jarvi AI Agent is the assistant inside Jarvi for daily recruiting work. Claude × Jarvi over MCP is the same data, accessed from your external Claude client.
Under three minutes for a single user. Workspace-wide rollout with scopes and SSO takes one short call with our team.
Three minutes to connect over MCP. Zero engineering. Your data stays in Jarvi.
The same Jarvi MCP endpoint powers every integration below. Pick the client your team already uses.
Already on Jarvi? Discover the Jarvi AI Agent, the in-product assistant powered by the same Claude model. New to Jarvi? See the Jarvi ATS or check pricing.
Native AI ATS, CRM and multi-channel outreach, in one workspace.
