Tutorial: Dedicated Networks MCP Server

In this tutorial you install the CAMARA Dedicated Networks MCP server, give it to an AI assistant, and ask the assistant, in plain language, to find a service area, reserve a dedicated network, give a device access to it and dismantle both again.

Read the READMEs first

Read the README of rt-camara-examples and the MCP server's own README before you start. They list what the server needs and every setting.

Repository	What it is	Used in
rt-camara-examples	The MCP server (`mcp-servers/dedicated-networks`), the management portal and the Insomnia collections	All sections

You want	Go to
The server installed	1. Install the server
A token for the sandbox	2. Get a token
An AI assistant to use the server	3. Give the server to an assistant
To book and release a network by asking	4. Ask the assistant
To call the tools from a script, or use OpenCode	Other ways to use the server

You need: git; Python 3.10 or newer with pip; an AI assistant that can use MCP servers, such as Claude Code; and test credentials for a CAMARA sandbox. The sandbox addresses and the credentials are provided to 5G-MAG members on request.

1. Install the server

git clone https://github.com/5G-MAG/rt-camara-examples.git
cd rt-camara-examples/mcp-servers/dedicated-networks
pip install -r requirements.txt

This installs mcp[cli] (the MCP framework), httpx (the HTTP client) and pydantic (input validation). To check that it starts, run it by hand:

python server.py

The server writes one line to standard error and then waits for a client:

[camara-mcp] Starting | API root: http://localhost:9091/ | Auth: NO TOKEN

Press Ctrl+C to stop it. An MCP client starts the server itself, from the configuration in section 3.

2. Get a token

The server calls the sandbox with a bearer token. Request one with your client ID and secret from the token endpoint you were given (<TOKEN_URL>).:

curl -s -X POST <TOKEN_URL> -d grant_type=client_credentials -d client_id=<CLIENT_ID> -d client_secret=<CLIENT_SECRET>

The answer is a JSON document; copy the value of access_token. It is valid for one hour (expires_in is 3600), so request a new one when it expires.

Keep the token and the secret private

Do not put them in a repository, a screenshot or a chat. The configuration file you edit next holds the token; keep it on your own machine.

3. Give the server to an assistant

An assistant starts the server from a configuration block. Save this as camara.mcp.json:

{
  "mcpServers": {
    "camara-dedicated-networks": {
      "command": "python",
      "args": [
        "/path/to/rt-camara-examples/mcp-servers/dedicated-networks/server.py"
      ],
      "env": {
        "CAMARA_API_ROOT": "<CAMARA_API_ROOT>",
        "CAMARA_ACCESS_TOKEN": "<the access_token you copied>"
      }
    }
  }
}

Replace the path with the real path of server.py, <CAMARA_API_ROOT> with the base address of your sandbox, and the token with yours. Then start Claude Code with the server:

claude --mcp-config camara.mcp.json

The server's 12 tools are now available to the assistant, with the prefix mcp__camara-dedicated-networks__. Check it with the first question of the next section: the assistant calls camara_list_profiles and lists the profiles.

4. Ask the assistant

Ask in plain language. The assistant chooses the tools; these are the questions that take you through a full cycle, with what to expect.

"What network profiles are available?" The assistant lists the profiles with their id, device limit, throughput and QoS profiles (camara_list_profiles).
"Find service areas at latitude 48.8467, longitude 2.2530." It lists the areas that cover the point, and a link for each to view it on a map (camara_retrieve_service_areas). Open a link to see the area on geojson.io.
"Create a dedicated network using the profile called profile1 in the Roland Garros Tennis Courts area, from 2026-10-20T17:00:00Z to 2026-10-20T18:00:00Z." Give the window as start and end times with a time zone, starting now or later. The assistant reserves the network and answers with its id and status REQUESTED (camara_create_network).
"What is the status of network <id>?" Within a few seconds the status moves to RESERVED or ACTIVATED; ask again if it still says REQUESTED (camara_get_network).
"Grant phone number +1234567890 access to network <id>." It creates the access, with status REQUESTED, and gives its id (camara_create_access).
"Is access <id> granted yet?" The status becomes GRANTED, or REJECTED with the reason (camara_get_access).
"List all my dedicated networks and their status." (camara_list_networks)

Replace each <id> with the id from the earlier answer, or let the assistant carry it over in the conversation.

Dismantle

Release what you booked, the accesses first:

"Revoke access <id>." (camara_delete_access)
"Delete network <id>." (camara_delete_network)

Then ask for the list of networks again: the answer is "No dedicated networks found."

An example conversation

This is a real conversation, held with Claude Code as the assistant against the sandbox, with the answers shortened. Your assistant will word its answers differently.

You: What network profiles are available?

The assistant calls camara_list_profiles and answers:

There are 6 network profiles available:

Name ID Max devices UL / DL QoS profiles (default in bold)
profile1 7e54f738-9e58-4509-9069-7023f03b22b6 100 100 / 100 Mbps QOS_A, QOS_B, QOS_C
profile2 3016629c-9c76-4d2c-8069-b50a24f0f62e 100,000 10 / 10 Mbps QOS_D, QOS_E, QOS_F
broadband 1e32608d-0e42-45e0-82a9-7c156972049c 1 1 / 10 Mbps eMBB
Medium OutsideBroadcast Production 914BA5A4-B1DE-404F-B7BB-D5B89FFDBE08 12 40 / 10 Mbps QoS_Ul20Mbps_Dl5Mbps, QoS_Ul10Mbps_Dl5Mbps, QoS_Bi1Mbps, QoS_RTC_64kbps
Small OutsideBroadcast Production 5961C65A-290D-40FE-951F-902B909B5C38 6 15 / 5 Mbps QoS_Ul10Mbps_Dl5Mbps, QoS_Bi5Mbps, QoS_Bi1Mbps, QoS_RTC_64kbps
MoJo/ENG Production 629CD271-4DF2-47B8-B646-D2559D5C6726 1 5 / 5 Mbps QoS_Bi5Mbps, QoS_Bi1Mbps

Name	ID	Max devices	UL / DL	QoS profiles (default in bold)
profile1	`7e54f738-9e58-4509-9069-7023f03b22b6`	100	100 / 100 Mbps	QOS_A, QOS_B, QOS_C
profile2	`3016629c-9c76-4d2c-8069-b50a24f0f62e`	100,000	10 / 10 Mbps	QOS_D, QOS_E, QOS_F
broadband	`1e32608d-0e42-45e0-82a9-7c156972049c`	1	1 / 10 Mbps	eMBB
Medium OutsideBroadcast Production	`914BA5A4-B1DE-404F-B7BB-D5B89FFDBE08`	12	40 / 10 Mbps	QoS_Ul20Mbps_Dl5Mbps, QoS_Ul10Mbps_Dl5Mbps, QoS_Bi1Mbps, QoS_RTC_64kbps
Small OutsideBroadcast Production	`5961C65A-290D-40FE-951F-902B909B5C38`	6	15 / 5 Mbps	QoS_Ul10Mbps_Dl5Mbps, QoS_Bi5Mbps, QoS_Bi1Mbps, QoS_RTC_64kbps
MoJo/ENG Production	`629CD271-4DF2-47B8-B646-D2559D5C6726`	1	5 / 5 Mbps	QoS_Bi5Mbps, QoS_Bi1Mbps

You: Find service areas at latitude 48.8467, longitude 2.2530.

The assistant calls camara_retrieve_service_areas and answers:

Two service areas cover that point (48.8467, 2.2530).

1. Roland Garros Tennis Courts (625b2d4b-4da7-4f07-9169-e60ffdf76671)

Circle, centre (48.845867, 2.253156), radius 1000 m

Network profiles: profile1 (7e54f738-...), profile2 (3016629c-...)

QoS profiles: QOS_S, 58e104f7-7b61-4603-9d0a-4813583c27e9

GeoJSON map

2. Grand Paris (4161b722-fa41-4e1d-933d-ed0aabb0fa08)

Polygon covering the Paris metropolitan region

Network profiles: broadband (1e32608d-...)

QoS profiles: eMBB

GeoJSON map

You: Create a dedicated network using the profile called profile1 in the Roland Garros Tennis Courts area, from 2026-10-20T17:07:16Z to 2026-10-20T18:07:16Z.

The assistant calls camara_create_network and answers:

The network request was submitted and is in REQUESTED status, so it's waiting for the provider to approve it.

Network ID: e90cdef4-67ee-4e34-9a16-3d1f7ae01a3e

Profile: profile1 (100 devices, 100/100 Mbps, QoS_A, QOS_B and QOS_C, default QOS_A)

Area: Roland Garros Tennis Courts

Service time: 2026-10-20T17:07:16Z to 2026-10-20T18:07:16Z

You: What is the status of that network?

The assistant calls camara_get_network and answers:

The network e90cdef4-67ee-4e34-9a16-3d1f7ae01a3e is still REQUESTED. The provider hasn't approved it yet.

You: Grant phone number +1234567890 access to that network.

The assistant calls camara_create_access and answers:

The access request for +1234567890 was submitted. It's in REQUESTED status and waiting for the provider to approve it.

Access ID: 40ad271a-44ff-4198-9e22-f235e2b03398

Network: e90cdef4-67ee-4e34-9a16-3d1f7ae01a3e

QoS profiles: all profile1 options (QOS_A, QOS_B, QOS_C), default QOS_A, because you didn't restrict them

You: Is that access granted yet?

The assistant calls camara_get_access and answers:

Yes, the access 40ad271a-44ff-4198-9e22-f235e2b03398 is now GRANTED for +1234567890.

You: Revoke that access.

The assistant calls camara_delete_access and answers:

I revoked access 40ad271a-44ff-4198-9e22-f235e2b03398, so +1234567890 can no longer use network e90cdef4-67ee-4e34-9a16-3d1f7ae01a3e.

You: Delete that network.

The assistant calls camara_delete_network and answers:

I deleted network e90cdef4-67ee-4e34-9a16-3d1f7ae01a3e, which cancels the reservation. Its device access was already revoked, so nothing is left from this session.

You: List all my dedicated networks.

The assistant calls camara_list_networks and answers:

You have no dedicated networks. That fits with the one we created having been deleted.

The server also gives the assistant four workflow prompts, dedicated_network_workflow, discover_profiles_and_areas, manage_device_access and teardown_network, which guide it through these steps. The parameters of all 12 tools and the recommended order of calls are in the tool reference.

Other ways to use the server

Call the tools from a script

Any MCP client can use the server, so you can try every tool without an assistant. Set CAMARA_API_ROOT and CAMARA_ACCESS_TOKEN in the environment, then, in the server's folder:

import asyncio, os
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

async def main():
    server = StdioServerParameters(command="python", args=["server.py"], env=dict(os.environ))
    async with stdio_client(server) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool("camara_list_profiles", {"params": {}})
            print("".join(c.text for c in result.content if hasattr(c, "text")))

asyncio.run(main())

Every tool takes its arguments in a single params object, for example {"params": {"atLocation": {"latitude": 48.8467, "longitude": 2.2530}}} for camara_retrieve_service_areas.

Use OpenCode

OpenCode reads MCP servers from an opencode.json in your project, or globally from ~/.config/opencode/opencode.json. Add:

{
  "mcp": {
    "camara-dedicated-networks": {
      "type": "local",
      "command": ["python", "/path/to/rt-camara-examples/mcp-servers/dedicated-networks/server.py"],
      "enabled": true,
      "environment": {
        "CAMARA_API_ROOT": "<CAMARA_API_ROOT>",
        "CAMARA_ACCESS_TOKEN": "<the access_token you copied>"
      }
    }
  }
}

Restart OpenCode. The tools then have the prefix camara-dedicated-networks_, for example camara-dedicated-networks_camara_list_profiles.

Run the server's tests

pip install -r requirements-dev.txt
python -m pytest

1. Install the server​

2. Get a token​

3. Give the server to an assistant​

4. Ask the assistant​

Dismantle​

Other ways to use the server​