Setup

setup — data-gouv-bys

You guide a new user through configuring the data-gouv-bys toolkit.

Prerequisites

• Claude Code CLI installed
• Internet access (the MCP and the API are remote)

Flow

1. Check data.gouv MCP connection

Try to call any tool from the official data.gouv MCP (e.g., search_datasets with query: "sirene").

If the MCP is not connected, help the user set it up:

claude mcp add --transport http datagouv https://mcp.data.gouv.fr/mcp

Explain: "Ce MCP est public et read-only, aucune clé API nécessaire. Il donne accès au catalogue data.gouv (datasets, ressources, API tierces, statistiques)."

After install, ask the user to restart Claude Code and confirm the connection by listing available tools (search_datasets, query_resource_data, etc.).

2. Test the public API Recherche d'entreprises

Run a sanity check via Bash:

curl -s "https://recherche-entreprises.api.gouv.fr/search?q=la+poste&per_page=1" -H "User-Agent: data-gouv-bys/0.1" | head -c 200

If the response contains "results" and a SIREN — green. If it 429s or fails, warn the user about the 7 req/s rate limit and propose adding a User-Agent header in all subsequent calls.

3. Optional — check the BYS Outbound Engine repo

Ask: "Tu as déjà cloné le repo bys_claude_outbound_engine à côté ? (pour le handoff plus tard)"

If yes → save the path in client.yaml (key: outbound_engine_path). If no → no problem, the user can clone it later when needed.

4. Collect client info

Ask the user, one question at a time:

1. Nom de la boîte (votre client / vous-même)
2. Site web
3. LinkedIn (URL)
4. Nom de l'expéditeur (qui signera les emails plus tard)
5. Email de l'expéditeur
6. Titre / fonction de l'expéditeur
7. Budget max en euros pour les futures campagnes (informatif)
8. Pays cibles (par défaut : France ; le toolkit data.gouv ne couvre que la France — si autres pays, prévenir que le handoff sortira juste le brief et que l'outbound engine prendra le relais via Lemlist People DB)

5. Create the client folder

Folder name : clients/<slug>_<YYYY-MM-DD>/ where <slug> is the company name lowercased and dash-separated (e.g. acme_2026-05-21).

Check the folder does NOT already exist before creating it. If it does, ask: "Un dossier clients/acme_2026-05-21/ existe déjà — tu veux le réutiliser, le dupliquer, ou en créer un autre ?"

6. Generate client.yaml

Write clients/<slug>_<YYYY-MM-DD>/client.yaml using the template at templates/client.yaml.example:

# clients/acme_2026-05-21/client.yaml
client:
  name: "ACME"
  website: "https://acme.com"
  linkedin: "https://linkedin.com/company/acme"
  countries:
    - "FR"

sender:
  name: "Marie Dupont"
  email: "marie@acme.com"
  title: "Head of Sales"

budget:
  max_eur: 5000

outbound_engine:
  path: "../bys_claude_outbound_engine"  # optional, only if cloned next to this repo

created_at: "2026-05-21"

7. Confirm and route

Display a short summary :

✓ MCP data.gouv connecté
✓ API Recherche d'entreprises joignable
✓ Dossier client créé : clients/acme_2026-05-21/
✓ client.yaml généré

Then route:

"Tout est prêt. Lance /data-gouv-bys:ciblage pour démarrer la méthode BYS (Discovery → CAB-P → 10 ciblages)."

Rules

• NEVER overwrite an existing client.yaml without explicit confirmation
• NEVER hardcode secrets — this plugin doesn't need any API key, but if the user has the outbound engine cloned, do not copy their .env.local here
• Always confirm the folder name before creating it
• Always test both the MCP AND the public API — they're complementary, both are needed downstream