Marvin's MCP server opens your recruiting database to your AI assistant: candidates, jobs, processes, missions, messages. In Claude, ChatGPT or Cursor, with your permissions. It's included in every plan, with no credit per request, and you set it up by pasting an address into your assistant's connectors. Your AI subscription stays yours.
Find five accountants available in November in my talent pool and enrol them in my “Accountancy Follow-up” sequence.
I don't have access to your talent pool or your sequences. Here's how to go about it:
No tool called. No data from your database.
search5 results
and 2 more profiles
Five profiles match in your talent pool. I'm preparing their enrolment: here's the preview before anything runs.
enroll_in_sequencedoneEnrol 5 candidates in “Accountancy Follow-up”First email: tomorrow, 9 AM. Sent from your account.
Done. All five candidates are enrolled, and the first email goes out tomorrow at 9 AM.
Demo, fictional profiles. Generic AI client: the look varies by assistant.
The tool names are those of Marvin's MCP server.
Installation
Not a line of code. The same address for everyone: your sign-in decides the account, the firm and the permissions.
https://mcp.marvins.ai/mcpOne address for every firm and every MCP-compatible assistant.
Paste the address, grant access in one click. The exact path varies from one assistant to another.
Works with Claude (Desktop, Web and Code), ChatGPT through connectors, Cursor, Goose, and any MCP-compatible assistant.
What your AI can do
Read, write, act: all of Marvin exposed to your assistant, with the same permissions as your account.
“Add a note to this contact's record.”
Marvin · createentity: note
Note added · Paul Berger“Call back after Thursday's committee.”“Move her to the interview stage.”
Marvin · advance_process_stage
Léa Garnier · Junior accountant“Who in my pool knows Kafka and stays under €600 a day?”
Marvin · searchentity: persons
Karim OualiSAP, Sage · €580 / day
Vincent TranSAP, Cegid · €560 / day“Start the sequence on these five profiles.”
Marvin · enroll_in_sequence
“Where are this month's timesheets?”
Marvin · searchentity: cras
“Send her a WhatsApp to confirm Thursday.”
Marvin · send_message
Hi Antoine, confirming our meeting on Thursday at 10 am.
preview · to confirm“How many placements this month?”
Marvin · count
Nothing outside your firm, nothing beyond your permissions.
Demo examples, fictitious profiles. Tool names are those of Marvin's MCP server.
Use cases
“Summarise my 12 active processes: who moved over the weekend, who's stuck, who should I call today?”
searchstale_entities
The assistant reads the records' history, spots processes with no activity for five days and ranks client follow-ups by priority.
“My client wants a senior Head of Finance in Paris. Pull eight profiles and start the ‘Exec FR’ sequence.”
searchenrich_person_from_linkedinenroll_in_sequence
Your pool cross-checked with LinkedIn enrichment, a pitch per profile, the multichannel sequence ready. You approve, you share with the client.
“Find the five best profiles for the accountant role at Orvane Consulting and draft a message for each.”
match_candidates_to_jobdraft_outreach_for_person
Draft 1 · CamilleHi Camille, your background in data…Draft 2 · MehdiHi Mehdi, your background in data…Draft 3 · LéaHi Léa, your background in data…A search of the pool, comparison with the role, one personalised message per candidate. You review, then approve sending.
Illustrative examples, fictional client and company.
Recruiter prompts
The server doesn't only offer tools: it also ships 8 prompts, instructions written for one specific task. You point to the candidate, process, job, mission or company; the assistant reads the record in Marvin and writes the result. They appear in the clients that support MCP prompts.
A client brief, ready to send.
Candidate briefTwo-line profile, career path, key skills, strengths, points of attention and a recommendation (go, hold or no-go).
/brief-candidateFrom a candidate recordAn interview debrief, structured.
Interview summaryContext, how it went, strengths, reservations, fit with the role, decision and next steps. The assistant stays factual and invents nothing./interview-summaryFrom a processWhere a job stands, stage by stage.
Pipeline recapSummary, candidates per stage, highlights, risks and the day's priorities, for your team stand-up or a client update./recap-pipelineFrom a jobA personalised follow-up, to review.
Candidate follow-upTwo variants, LinkedIn and email, with a note recalling the assumptions made. The prompt drafts: it sends nothing./relance-candidatFrom a candidate recordA sourcing plan for the job.
Sourcing planNeeds summary, three target personas, LinkedIn queries, estimated volume against the remaining quota, channels and points of attention./sourcing-planFrom a jobA mission review, month by month.
Mission reviewScope, activity reported in the timesheets, highlights drawn from the notes and the timeline, points of attention and a suggested next step. Rates stay out of the text meant for the client./bilan-missionFrom a missionA client meeting prepared in one request.
Meeting preparationRelationship history, open jobs and processes, notes, latest exchanges, topics to raise, questions to ask and the next step to propose./prepa-rendez-vousFrom a company (and a contact, if needed)A market map, built on your own database.
Market mapTarget companies, signals (open jobs, latest exchange, client status), profiles already in your database and where to start. What your database doesn't know is flagged, not invented./cartographie-marcheFrom a job or a sectorOrder of magnitude
What this could mean in tasks done through the assistant, using cautious assumptions to adapt to your team.
Illustrative estimate, not a measurement
Calculation: 40 × 50% × 4 × 21 = 1,680
A task is a request carried through with the assistant: a brief written, a process stage advanced, a search run, a note saved.
No usage figures are published here: this number comes from the calculation above, not from a real agency. Your own result will depend on your team size and adoption.
Why this one
The price
The MCP server is part of every Marvin plan. The AI runs in your assistant, on your subscription: Marvin rebills nothing.
On your database, from your assistant.
Copy the address, grant access.
No meter, no surcharge.
Included in both plans. Pricing
Security and compliance
Every recruiter signs in with their own account.
OAuth 2.1, per userNo shared token, no dormant access.No way out of your firm.
Strict isolationEvery MCP request is filtered on your firm and your roles.Hosted in France, erasable on request.
GDPRScaleway, in Paris. Traffic over TLS, secrets and tokens encrypted with AES-256-GCM. The right to erasure applies to the MCP channel like everywhere else. SecurityEvery action stays logged, under your name.
Your permissions, your trailThe assistant does nothing your account couldn't do. Responsible AIFrequently asked questions
MCP (Model Context Protocol) is an open standard, published by Anthropic, that lets an AI assistant connect to your business tools. Marvin's MCP server gives your assistant controlled access to your database: no more copy-pasting your data into the chat.
Claude (Desktop, Web with custom connectors, and Code), ChatGPT through connectors, Cursor, Goose, and any MCP-compatible assistant.
The assistant only receives the data you ask it for, for the length of the conversation, under the retention rules of your account with that provider. Your database stays hosted by Marvin, in France, with its secrets encrypted.
No. Copy the address https://mcp.marvins.ai/mcp, paste it into your assistant's connectors, grant access in one click. Under two minutes.
Nothing extra: it's included in every Marvin plan, with no credit per request. Your AI subscription (Claude, ChatGPT, Cursor…) stays yours.
Not by default. Sensitive actions, such as sending a message or enrolling in a sequence, go through a preview first: you confirm, then Marvin runs it.
Turn on MCP for your data.
Available on every plan.A 30-minute demo, on your own data, in the assistant you already use.
Updated