Connect an assistant
Claude, ChatGPT, Le Chat and Claude Code use the same address,
https://mcp.madata.africa/mcp, with no key to paste: you pick the
workspace when signing in. Tools configured by file use an
MCP key.
01Before you start
- Mia active on the workspace. Without a Mia subscription, MADATA refuses the authorization. Add it from the portal Marketplace.
- A MADATA account with Mia. Owner, administrator or a user with Mia enabled. The assistant acts with your permissions, no more, no less.
- A compatible assistant. Custom connectors depend on the assistant's plan.
For clients, the illustrated guide in the portal walks through these steps with screenshots, also available as a PDF (in French).
02 Claude
On claude.ai or in the desktop app.
- Open Customize → Connectors → Add → Add custom connector.
- Name: Madata. URL:
https://mcp.madata.africa/mcp. Click Continue. Leave the OAuth client ID fields empty. - The MADATA page opens: sign in, pick the workspace, then Authorize. If it does not open, click Connect next to Madata.
- In a conversation, + button then Connectors: switch on Madata.
03 ChatGPT
On chatgpt.com or in the desktop app.
- Open Plugins → Add → Add custom MCP server.
- Name: Madata. Server URL:
https://mcp.madata.africa/mcp. Authentication: OAuth. - Tick I understand and want to continue, then Create as plugin.
- The MADATA page opens: sign in, pick the workspace, then Authorize.
To comply with the OpenAI directory rules,
the gateway removes 12 tools for ChatGPT: query_records,
aggregate_records, export_records,
create_record, update_records, action_record,
create_payment_link, send_whatsapp_payment_link,
schedule_task, install_routine_recipe,
create_mission, check_app_access. They do not appear in
tools/list and calling them returns -32602. Claude and MCP
keys keep the whole catalog.
Or through a custom GPT Action
Import the workspace's OpenAPI schema:
https://<workspace>.madata.app/mia/mcp/openapi.json
Under Authentication, choose API Key, type Bearer, and paste an MCP key. See OpenAPI spec.
04 Mistral Le Chat
- Open Le Chat → Settings → Connectors → Add a connector (MCP).
- Paste
https://mcp.madata.africa/mcp, authentication OAuth. - Click Connect, sign in, pick the workspace, then Authorize.
If your version asks for a key instead of OAuth, create an MCP key and paste it
as Bearer with the workspace address.
05 Claude Code
Add the server once in the terminal, then type /mcp, choose
madata and Authenticate: the MADATA page opens in
your browser.
claude mcp add --transport http madata https://mcp.madata.africa/mcp
06 Gemini CLI
With an MCP key, read-only to start with. Restart Gemini CLI: the MADATA tools show up.
{
"mcpServers": {
"madata": {
"httpUrl": "https://<workspace>.madata.app/mia/mcp",
"headers": { "Authorization": "Bearer mia_<your_key>" }
}
}
}
07Cursor and VS Code
Same principle: an MCP key and the workspace address.
{
"mcpServers": {
"madata": {
"url": "https://<workspace>.madata.app/mia/mcp",
"headers": { "Authorization": "Bearer mia_<your_key>" }
}
}
}
{
"servers": {
"madata": {
"type": "http",
"url": "https://<workspace>.madata.app/mia/mcp",
"headers": { "Authorization": "Bearer mia_<your_key>" }
}
}
}
08Your scripts: curl, Python, any language
The server is a standard JSON-RPC 2.0 endpoint. Three calls are enough to
get going: initialize, tools/list,
tools/call.
BASE="https://<workspace>.madata.app/mia/mcp"
AUTH="Authorization: Bearer mia_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
curl -s "$BASE" -H "$AUTH" -H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize",
"params":{"protocolVersion":"2025-06-18","capabilities":{},
"clientInfo":{"name":"cli","version":"1"}}}'
curl -s "$BASE" -H "$AUTH" -H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
"params":{"name":"list_overdue_followups","arguments":{}}}'
import httpx
MCP_URL = "https://<workspace>.madata.app/mia/mcp"
API_KEY = "mia_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
def mcp_call(method, params=None):
resp = httpx.post(
MCP_URL,
json={"jsonrpc": "2.0", "id": 1, "method": method, "params": params or {}},
headers={"Authorization": f"Bearer {API_KEY}"},
timeout=30,
)
return resp.json().get("result")
tools = mcp_call("tools/list")["tools"]
kpi = mcp_call("tools/call", {"name": "get_kpi", "arguments": {"period": "month"}})
09Stay in control
In the MADATA portal, Mia page, MCP tab, Connections and keys card:
- Read only or Read + write: per connection.
- Cut access: revoking acts at once, deleting erases the connection.
- Your MADATA permissions: the assistant only sees and does what your account allows.
- Before writing: your assistant asks for your approval, depending on its tool permission settings.
10Troubleshooting
| Symptom | Cause and fix |
|---|---|
| "AI assistants are not enabled on this workspace" | The workspace has no active Mia subscription. Add Mia from the Marketplace, then connect again. |
| Claude: "the connection succeeded, but the server refused the request" | Same cause: Mia subscription stopped or Mia turned off. Reactivate Mia; access comes back with nothing to redo. |
| The assistant does not find Madata | In the conversation, + button then Connectors: check that the Madata switch is on. |
| The assistant reads but cannot create anything | The connection or your Mia access is read only. Switch it to Read + write in Connections and keys, or ask your administrator. |
| The wrong workspace is linked | Disconnect Madata in the assistant, connect again and pick the right workspace. |
401 with a mia_… key on mcp.madata.africa | The gateway does not accept portal keys. Use the workspace address. |