Connect an agent

Wire any agent to Enjab RAG over MCP so it shares the same operational knowledge as every other Enjab agent.

Enjab RAG ships a native MCP server (Streamable HTTP, JSON-RPC). Point any MCP-capable agent at it and it gets two tools, search and fetch, no SDK or glue code required. This is the way agents connect.

Endpoint: https://rag.enjab.ae/api/mcp
Auth: Authorization: Bearer enjabrag_… (an API key)
Protocol: MCP 2024-11-05, transport http

Mint an API key

Sign in at rag.enjab.ae/dash/keys and create a key for the agent. A read-only key is all an agent needs to search and fetch. Copy it once, only its hash is stored. See Authentication.

Add the MCP server

Claude Code (CLI):

claude mcp add --transport http enjab-rag https://rag.enjab.ae/api/mcp \
  --header "Authorization: Bearer enjabrag_YOUR_KEY"

Generic MCP client (mcp.json / config):

{
  "mcpServers": {
    "enjab-rag": {
      "type": "http",
      "url": "https://rag.enjab.ae/api/mcp",
      "headers": { "Authorization": "Bearer enjabrag_YOUR_KEY" }
    }
  }
}

Use the tools

The agent now has:

search — { query: string, limit?: number, collection?: string }. Returns ranked passages with each one's source title, summary, and metadata. The agent calls this to ground answers about how the clinics operate.
fetch — { id: string }. Pass the document_id from a search result to read that whole source back (title, full extracted text, summary, tags, metadata).

Tell the agent, in its system prompt, to search Enjab RAG before answering operational questions so it grounds on shared knowledge instead of guessing.

Hand this to your agent

Drop this into the agent you are building so it connects itself:

Connect to Enjab RAG, the shared knowledge base for Enjab's agents. It is an MCP
server at https://rag.enjab.ae/api/mcp (transport: http) authenticated with a Bearer
API key (format enjabrag_...). Add it as an MCP server with the header
"Authorization: Bearer <key>". It exposes two tools: `search` (query the shared
operational knowledge) and `fetch` (read a full document by id). Before answering any
question about how the clinics operate, call `search` first and ground the answer in
the results. Read https://developers.enjab.ae/docs/enjab-rag for the full reference.

Raw protocol (custom clients)

The endpoint is plain JSON-RPC over HTTP POST. Initialize, list tools, then call one:

# tools/list
curl -s https://rag.enjab.ae/api/mcp \
  -H "Authorization: Bearer enjabrag_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# tools/call -> search
curl -s https://rag.enjab.ae/api/mcp \
  -H "Authorization: Bearer enjabrag_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"search","arguments":{"query":"what time do the clinics open?"}}}'

tools/call returns a content array (a human-readable text block) plus structuredContent (the raw results array, same shape as the REST search response). Supported methods: initialize, notifications/initialized, ping, tools/list, tools/call.

The MCP server is read-oriented (search + fetch). To add knowledge programmatically, use the write-scoped /api/ingest endpoint.

Mint an API key

Add the MCP server

Use the tools

Hand this to your agent

Raw protocol (custom clients)

On this page