This feature is currently in closed-beta. If you are interested in participating in the beta program, please contact your Customer Success Manager (CSM).
Use this article to install Cint MCP in Claude Code, including the Code view in the Claude desktop app. If you use the Claude app on desktop or web, see Getting started with Cint MCP in Claude instead.
Setup in Claude Code differs from the Claude app in one way: Add the connector with a command or a configuration file instead of the Claude interface.
In Claude Code, the install is the same two steps as in the Claude app: add the server, then sign in.
Once Cint MCP is installed, using it is the same in both. For the next steps, see Run your first feasibility check with Cint MCP and How to use Cint MCP.
This article covers how to add the Cint server, sign in, remove an old skill, set per-tool permissions, and share the connector with a team.
Add the server
Add the Cint server with the command below. For the current syntax and the other options, follow Claude Code's instructions → Connect Claude Code to tools via MCP
claude mcp add --transport htt cint-mcp https://api.cint.com/mcp --scope user
Part of the command | What it does |
|---|---|
| Connects to the server over HTTP |
| The name of the server. Use this exact name. |
| The address of the Cint server |
| Makes the server available in all of your projects |
The command does not include a client ID, client secret, or scopes. The Cint server registers itself with Claude Code when you sign in. Earlier versions of this guide used a --client-id flag, which is no longer needed.
Use the exact server name
cint-mcp
Claude Code adds the server name to the front of every tool the server provides. With this name, the tools appear as mcp__cint-mcp__exchange_lists_accounts. For example: The permission rules in thePer-tool permissions section match on this prefix.
If you use a different name, such as cint, Cint-MCP, or cint_mcp, Cint MCP still works. However, none of the permission rules in the Per-tool permissionssection match your tools, and Claude Code asks you to approve every call, including reads.
After you add the server, Claude Code lists it as Needs authentication and waits. Sign-in does not start automatically.
Sign in to Cint
Signing in works the same way as in the Claude app. Claude acts as you, with your Cint permissions, on the accounts your login reaches. The connector holds no credentials of its own.
You do not need a client ID or client secret. The Cint server registers itself with Claude Code when you sign in (this is called dynamic client registration). Nothing needs to be registered in advance.
To sign in, use one of the following methods:
Option 1: From the terminal
claude mcp login cint-mcp
Option 2: From a Claude Code session
Start a session and run
/mcp.Select the
cint-mcpentry marked "Needs authentication" and authorize it.
Both methods open a browser window. Sign in with your own Cint credentials, then return to the terminal.
What signing in does
Use the steps outlined in the Cint MCP Install and Connect article to sign in.
Claude acts as you, with your Cint permissions, on the accounts your login reaches, and the connector holds no credentials of its own.
You do not need a client ID or client secret. The Cint server registers itself with Claude Code when you sign in (this is called dynamic client registration). Nothing needs to be registered in advance.
How long sign-in lasts
Claude Code stores your token in your system keychain and refreshes it automatically, so you sign in once.
Claude Code and the Claude app sign in separately
Signing in to one does not sign you in to the other. This includes the Code view in the Claude desktop app, which is Claude Code. If Cint MCP works in the Claude app but not in Claude Code, sign in here.
Check that sign-in worked
Start a new session and ask “check my Cint connection.” Claude should report the Cint account it is working in. Confirm it worked in the Verify your setup section.
Check the skill after you install it by using the steps in the Verify your setup section.
Claude Code now gets its Cint guidance from the server. A copy of the cint-demand-mcp skill left over from an earlier round of the beta conflicts with it, so delete it with the command below. If you never installed it, there is nothing to do.
rm -rf ~/.claude/skills/cint-demand-mpc
Verify your setup
If Cint MCP isn't working in Claude Code, find your symptom in the table.
Symptom | Cause | Fix |
|---|---|---|
| The server was added in a different scope, or the | Re-run the Add the server command with See Set up Cint MCP in the Claude Code CLI for more information. |
The server shows as failed or disconnected. | The token expired, or sign-in never completed. | Run |
Sign-in completes, but no accounts come back. | Your credentials are valid, but they are not attached to an account with Exchange access. | Contact your Cint representative. |
Tools are listed twice, and approvals appear twice. | An old | Run |
Tools work, but every call prompts for approval. | Your permission patterns still use the old | Replace them with the rules in the Per-tool permissions section. See Set up Cint MCP in the Claude Code CLI for more information. |
Updating
There is nothing to update. Changes arrive from the Cint server. Start a new session to pick them up.
Per-tool permissions
Cint tool names start with the product, followed by the verb, such as exchange_check_feasibility and exchange_update_cpi. Within the Exchange, reads begin with get_, list_, check_, search_, load_, count_, find_, or diagnose_. Writes begin with create_, update_, transition_, or request_. You can use this split to allow the reads outright and hold the writes for approval, with no per-tool maintenance as new tools are added.
Cint tool names divide by verb. Reads begin with get_, list_, check_, search_, load_, or count_. Writes begin with create_, update_, launch_, pause_, resume_, or close_. You can use this split to allow the reads outright and hold the writes for approval, with no per-tool maintenance as new tools are added.
For where these settings go and the full rule syntax, follow Claude Code's instructions:
The new rules:
{
"permissions": {
"allow": [
"mcp__cint-mcp__exchange_get_*",
"mcp__cint-mcp__exchange_list_*",
"mcp__cint-mcp__exchange_check_*",
"mcp__cint-mcp__exchange_search_*",
"mcp__cint-mcp__exchange_load_*",
"mcp__cint-mcp__exchange_count_*",
"mcp__cint-mcp__exchange_find_*",
"mcp__cint-mcp__exchange_diagnose_*"
],
"ask": [
"mcp__cint-mcp__exchange_create_*",
"mcp__cint-mcp__exchange_update_*",
"mcp__cint-mcp__exchange_transition_*",
"mcp__cint-mcp__exchange_request_*"
]
}
}Updating an existing settings file
Replace the old rules. Both the server prefix and the tool names have changed. The prefix mcp__Cint_Exchange_MCP__ is now mcp__cint-mcp__, and tool names are now prefixed exchange_.
Old rules do not produce an error. They match nothing, so every call falls through to a prompt, or, worse, to whatever your default is.
The only change is the prefix: mcp__Cint_Exchange_MCP__ becomes mcp__cint-mcp__. The patterns are otherwise identical.
Old rules do not produce an error. They match nothing, so every call falls through to a prompt, or, worse, to whatever your default is.
This is also why the steps within Add a Server nsists on the exact server name. The prefix is derived from it.
Do not use blanket permission-bypass modes
Claude Code offers modes that skip permission prompts entirely. Do not use them with this server.
The reads are harmless. The writes are not: exchange_transition_target_groups puts a target group into field or ends collection, and exchange_update_cpi changes what you pay per complete. These actions affect live commercial work with real budget at stake, in an account you may not have looked at recently. A bypass mode removes the last point at which a wrong account or a wrong figure is cheap to catch.
If prompts are slowing you down, widen the allow list in thePer-tool permissions section rather than switching off the approval mechanism. Allowing every read tool removes the great majority of prompts and leaves every state change in front of you.
Sharing the connector with a team
To make the server available to everyone who works in a repository, commit a .mcp.json file at the root of the repository:
{
"mcpServers": {
"cint-mcp": {
"type": "http",
"url": "https://api.cint.com/mcp"
}
}
}"type": "http" is required. Claude Code skips an entry that has a url but no type, and it does not tell you.
Each person still runs claude mcp login cint-mcp once, with their own credentials. The committed file only says where the server is. It does not give anyone access to anything.
For the config file format and scope behavior, follow Claude Code's instructions → Connect Claude Code to tools via MCP
The Code view inside the Claude desktop app
The Code view in the Claude desktop app is Claude Code, and it authorizes separately from the rest of the app. Signing in to the connector in the Claude app does not carry over to the Code view.
To authorize it:
Open an interactive session in the Code view.
Run
/mcpAuthorize the server.
Until you do, Claude will say that the server needs authorization and that the session cannot run the sign-in flow itself. Please note, this message is expected, and it does not mean something is broken.
Direct questions about the Exchange itself to your Cint contact. This includes what is planned, how a commercial term works, and whether your account can be configured differently.