> ## Documentation Index
> Fetch the complete documentation index at: https://docs.yambr.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Custom MCP clients

> Talk to the Computer Use Server from any HTTP client that speaks MCP JSON-RPC.

The server speaks standard **MCP over Streamable HTTP**. Any HTTP client can drive it; you just need to handle the JSON-RPC envelope, session id, and SSE framing.

## Initialize

```bash theme={null}
curl -sD - -X POST "http://localhost:8081/mcp" \
  -H "Authorization: Bearer $MCP_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-Chat-Id: my-session" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2024-11-05",
      "capabilities": {},
      "clientInfo": {"name": "my-client", "version": "1.0"}
    }
  }'
```

Save the `Mcp-Session-Id` response header — you'll pass it on every subsequent call.

## List tools

```bash theme={null}
curl -s -X POST "http://localhost:8081/mcp" \
  -H "Authorization: Bearer $MCP_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: $SESSION_ID" \
  -H "X-Chat-Id: my-session" \
  -d '{"jsonrpc": "2.0", "id": 2, "method": "tools/list"}'
```

## Call a tool

```bash theme={null}
curl -s -X POST "http://localhost:8081/mcp" \
  -H "Authorization: Bearer $MCP_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: $SESSION_ID" \
  -H "X-Chat-Id: my-session" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "bash_tool",
      "arguments": {"command": "echo Hello", "description": "test"}
    }
  }'
```

## List / read resources

Uploaded files show up as `resources/list` entries with URIs of the form `file://uploads/{chat_id}/{encoded-rel-path}`. See [MCP methods → resources](/api-reference/mcp/resources).

## Dynamic system prompt

Non-MCP clients can fetch the per-session system prompt over plain HTTP:

```bash theme={null}
curl "http://localhost:8081/system-prompt?chat_id=my-session&user_email=user@example.com"
```

Stream it into the model's system message. The response carries an `X-Public-Base-URL` header with the server's public URL — use that when rewriting file links.

## Python example (mcp SDK)

```python theme={null}
from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async def main():
    async with streamablehttp_client(
        url="http://localhost:8081/mcp",
        headers={
            "Authorization": "Bearer " + MCP_API_KEY,
            "X-Chat-Id": "my-session",
        },
    ) as (read, write, _):
        async with ClientSession(read, write) as session:
            await session.initialize()
            tools = await session.list_tools()
            result = await session.call_tool(
                "bash_tool",
                {"command": "echo Hello", "description": "test"},
            )
            print(result)
```

## Managed Yambr

For managed Yambr, the same MCP endpoint is public at `https://api.yambr.com/mcp/computer_use` with `Authorization: Bearer sk-yambr-...`. Use the exact same `initialize` / `tools/list` / `tools/call` / `resources/*` calls shown above — just change the URL and the Bearer token. Your LLM traffic stays on your own provider (OpenAI, Anthropic, whatever); Yambr only provides the Computer Use tools. See [LiteLLM gateway](/platform/litellm) and [Access model](/platform/access-model).

## See also

* [MCP API reference](/api-reference/introduction) — tool list, headers, dynamic endpoints
* [System prompt](/reference/system-prompt) — six MCP-native tiers