Skip to content
For AI agents: the complete documentation index is available at llms.txt; this page is also available as Markdown at index.md.

Work with MCP servers

Cohesivo comes with a built-in MCP server called management, which exposes a set of built-in tools to AI agents.

The server is available under the /mcp/management path, for example:

1
https://example.cohesivo.app/mcp/management

The server uses the Streamable HTTP transport.

Connect to the MCP server

The MCP server accepts the same OAuth 2.0 access tokens as the REST API. Send the token in the Authorization header of every request:

1
Authorization: Bearer <access_token>

For information about how to get a token, see REST API authentication.

Unlike the REST API, the MCP server has no anonymous access. Requests without a valid token respond with 401 Unauthorized:

1
2
3
{
    "error": "MCP endpoints require authentication."
}

Use the built-in tools

Once an agent connects, it discovers the available tools. Based on the prompt you provide to the agent, the agent decides which tools to call.

The built-in tools let an agent work with content types, field definitions, content type groups, translations, and SEO metadata. For example, you can ask the agent to create a content type with a set of fields, to list the content items that have no translation into a given language, or to point out the content items that are missing a meta title.

Because the tools act through the API, everything the agent can do is limited by the permissions of the user that the access token represents. With the client credentials flow, this is the service account.

Test the MCP server with Copilot CLI or Claude Code

You can test your MCP server with Copilot CLI or Claude Code, as illustrated here, or with any other agent or interface.

Add MCP server to agent CLI

To add an MCP server to your agent, you need to request a JWT token, then configure the MCP server in your agent as a Streamable HTTP with a Authentication: Bearer <token> header.

The following script uses curl to request a JWT token, then creates or updates the MCP server configuration. It works with Copilot CLI and Claude Code CLI.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
#!/bin/bash
set -e

# Adapt to your test case
baseUrl='http://localhost'
serverUrl="$baseUrl/mcp/example"
serverName='ibexa-example'
username='ibexa-example'
password='Ibexa-3xample'
agentCli='copilot'
#agentCli='claude'

jwtToken=$(curl -s -X 'POST' \
  "$baseUrl/api/ibexa/v2/user/token/jwt" \
  -H 'Content-Type: application/vnd.ibexa.api.JWTInput+json' \
  -H 'Accept: application/vnd.ibexa.api.JWT+json' \
  -d "{
        \"JWTInput\": {
          \"_media-type\": \"application/vnd.ibexa.api.JWTInput+json\",
          \"username\": \"$username\",
          \"password\": \"$password\"
        }
      }" | jq -r .JWT.token)

if [[ -n $($agentCli mcp list | grep -E "(^| )$serverName\W") ]]
then
  $agentCli mcp remove "$serverName"
fi

$agentCli mcp add --transport http \
  "$serverName" "$serverUrl" \
  --header "Authorization: Bearer $jwtToken"

When the agent can't connect to the MCP server, re-run the script.

Rate limits

To prevent abuse, MCP server calls are rate limited.

Responses carry the current state of the quota:

Header Description
X-RateLimit-Limit Number of requests permitted in the window.
X-RateLimit-Remaining Number of requests still available.
X-RateLimit-Reset Unix timestamp at which the window resets.

When you exceed the quota, the request responds with 429 Too Many Requests and a Retry-After header that tells you how many seconds to wait.