> For the complete documentation index, see [llms.txt](https://docs.nected.ai/nected-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.nected.ai/nected-docs/ai-assistant-copilot/nected-mcp-server.md).

# Nected MCP Server

Connect Nected to AI Agents via MCP

The **Nected MCP Server** brings your Nected workspace into the AI agents your team already uses, such as Claude or ChatGPT. Once Nected is connected, the agent can find, build, test, publish, run, and audit your rules and workflows from a conversation, which makes your complete decisioning flow agentic.

Your decision logic stays in Nected. The agent uses Nected to make decisions instead of improvising them. The result is the same predictable, governed outcome every time, whether the decision is triggered by your application, a schedule, or an AI agent.

New to MCP? See the official [Model Context Protocol introduction](https://modelcontextprotocol.io/docs/getting-started/intro).

{% hint style="info" %}
**Looking to bring an external MCP server into Nected?** This page is about connecting Nected *to* your AI agents. To connect your own or a third-party MCP server *inside* Nected (for example in an AI Agent node), see [MCP Server integration](https://docs.nected.ai/nected-docs/integrations/integrations-libraries/mcp-server).
{% endhint %}

***

## Why connect Nected to your AI agents

* **Make decision flows agentic.** Let AI agents design, update, test, and run your rules and workflows through plain-language requests, while Nected stays the system that executes them.
* **Get deterministic, predictable decisions.** The agent decides *when* a decision is needed. Nected decides *what* the outcome is. The same input always produces the same result, based on rules you can read, test, and version.
* **Add guardrails to your AI agents.** Instead of letting a model guess eligibility, pricing, limits, or approvals, route those decisions through Nected rules. Agents stay within the business logic you've defined and published.
* **Keep governance with Nected RBAC.** The agent acts as the signed-in user and can never do more than that user's Nected role allows. If you can't publish a rule in the console, the agent can't publish it either.
* **Work securely as an organization.** Access is scoped to one workspace per connection, sign-in uses OAuth (no API keys to share), secrets stay masked, changes are saved as drafts, runs default to staging, and every execution is recorded in the audit log.
* **Nothing to install.** Nected MCP is a hosted server. You only add its URL to your AI agent.

***

## Nected MCP configuration

Use this endpoint to connect Nected to any MCP-compatible AI agent:

```
https://api.nected.ai/mcp
```

{% hint style="warning" %}
This URL is an MCP endpoint, not a web page. Add it to your AI agent. If you open it in a browser, you'll only see a "session expired" error.
{% endhint %}

| Setting        | Value                                                 |
| -------------- | ----------------------------------------------------- |
| Transport      | Streamable HTTP (remote MCP server)                   |
| Authentication | OAuth 2.0. Sign in with your Nected account.          |
| Scope          | One workspace per connection, chosen when you sign in |
| Access         | Based on your Nected role (RBAC) in that workspace    |

### Available tools

After you connect, the agent can use these tools. It picks the right tool from your request, so you don't need to call them by name.

<table><thead><tr><th width="149.6796875">Area</th><th width="221.04296875">Tool</th><th>What it does</th></tr></thead><tbody><tr><td>Rules</td><td><code>listRule</code></td><td>List rules, filtered by name, type, status, or version.</td></tr><tr><td></td><td><code>getSimpleRule</code></td><td>Read a simple rule.</td></tr><tr><td></td><td><code>getDecisionTable</code></td><td>Read a decision table.</td></tr><tr><td></td><td><code>getRuleset</code></td><td>Read a ruleset.</td></tr><tr><td></td><td><code>createSimpleRule</code></td><td>Create a simple rule as a draft.</td></tr><tr><td></td><td><code>createDecisionTable</code></td><td>Create a decision table as a draft.</td></tr><tr><td></td><td><code>createRuleset</code></td><td>Create a ruleset that chains existing rules, as a draft.</td></tr><tr><td></td><td><code>updateSimpleRule</code></td><td>Update a draft simple rule.</td></tr><tr><td></td><td><code>updateDecisionTable</code></td><td>Update a draft decision table.</td></tr><tr><td></td><td><code>updateRuleset</code></td><td>Update a draft ruleset.</td></tr><tr><td></td><td><code>testRule</code></td><td>Test a rule with sample inputs, without publishing it.</td></tr><tr><td></td><td><code>publishRule</code></td><td>Publish the draft as the live version.</td></tr><tr><td>Workflows</td><td><code>listWorkflow</code></td><td>List workflows, filtered by name, status, or version.</td></tr><tr><td></td><td><code>getWorkflow</code></td><td>Read a workflow and its nodes.</td></tr><tr><td></td><td><code>createWorkflow</code></td><td>Create a workflow as a draft.</td></tr><tr><td></td><td><code>updateWorkflow</code></td><td>Update a draft workflow.</td></tr><tr><td></td><td><code>testWorkflow</code></td><td>Test a workflow with sample inputs, without publishing it.</td></tr><tr><td></td><td><code>publishWorkflow</code></td><td>Publish the draft as the live version.</td></tr><tr><td>Execution</td><td><code>execute</code></td><td>Run a rule or workflow in staging or production and return the result.</td></tr><tr><td>Audit logs</td><td><code>listExecution</code></td><td>Search execution logs by entity, status, environment, trigger, date, or input/output values.</td></tr><tr><td></td><td><code>getExecution</code></td><td>Read one execution: input, output, error message, and per-node results.</td></tr><tr><td>Workspace resources</td><td><code>listIntegration</code></td><td>List integrations (databases, APIs, spreadsheets, MCP servers, and more).</td></tr><tr><td></td><td><code>getIntegration</code></td><td>Read an integration's configuration. Secrets are masked.</td></tr><tr><td></td><td><code>listDataSource</code></td><td>List datasources.</td></tr><tr><td></td><td><code>getDataSource</code></td><td>Read a datasource's query, parameters, and returned fields.</td></tr><tr><td></td><td><code>listAttribute</code></td><td>List attribute libraries.</td></tr><tr><td></td><td><code>getAttribute</code></td><td>Read an attribute library and its attributes.</td></tr><tr><td></td><td><code>listDocument</code></td><td>List uploaded documents.</td></tr><tr><td></td><td><code>getDocument</code></td><td>Read a document's text.</td></tr><tr><td></td><td><code>listVariables</code></td><td>List global variables. Secret values are hidden.</td></tr><tr><td></td><td><code>listSystemVariables</code></td><td>List built-in system variables.</td></tr><tr><td>Authoring guides</td><td><code>listSkill</code></td><td>List Nected's built-in guides for building rules and workflows.</td></tr><tr><td></td><td><code>getSkill</code></td><td>Read a guide, so the agent builds entities the way Nected expects.</td></tr></tbody></table>

There are no tools to create or edit workspace resources, so manage integrations, datasources, attribute libraries, documents, and variables in the Nected console.

***

## How the agent works with your workspace

The agent follows the same lifecycle as the Nected console:

1. **Describe** what you need in plain language.
2. **Create or update.** The agent saves new and changed rules and workflows as **drafts**. Your live versions don't change.
3. **Test.** The agent tests the draft with sample inputs and shows you the results.
4. **Review.** Check the summary and test results before going further.
5. **Publish.** When you approve, the agent publishes the draft as the live version.
6. **Run and audit.** The agent runs the live version, in staging by default, and every run is recorded in the audit log.

{% hint style="info" %}
Before a large change, a publish, or a production run, ask the agent to list exactly what it will change or run. Approve only once the plan looks right.
{% endhint %}

For example, a team can build a ticket management and escalation system with their AI agent. The agent builds and tests the workflow in Nected, the results loop back to the agent, and the team publishes the workflow once it's ready.

<figure><img src="https://4290782554-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLg716fCfV8IUwXQygkTG%2Fuploads%2FHJK1kM89Z1nkp5Clbh4i%2Fnected-mcp-agentic-flow.png?alt=media&amp;token=cf708b5f-f5f2-4575-a43e-dcf60b1f826f" alt=""><figcaption></figcaption></figure>

* **Your team and AI agent.** The team submits its PRD to the AI agent. The agent reads the requirements, builds the Ticket Escalation workflow in Nected through MCP, and tests it with sample tickets.
* **Nected.** The workflow handles each ticket in multiple steps. It fetches the customer and SLA data, and Priority Rules set the priority and SLA the same way every time. An AI Agent node then assesses the ticket, and a switch node either sends it to human review, as a feedback loop, or approves it automatically.
* **Feedback and loop.** Test results and execution logs go back to the agent, which gives the team a summary of whether to publish the workflow or fine-tune it further. To fine-tune, the agent updates the draft and tests again. When it's ready, the team publishes the workflow from the Nected UI.

### Limitations

* **One workspace per connection.** To work in another workspace, switch workspaces.
* **No deletes.** The agent can't delete rules, workflows, or any other entity.
* **Workspace resources can't be edited.** Integrations, datasources, attribute libraries, documents, and variables can be read but not created or changed.
* **Drafts first.** New and updated rules and workflows are always saved as drafts, and only reach live callers when they're published.
* **Your role applies.** The agent can only do what your Nected role allows in that workspace.
* **Production runs are real.** Running in production uses your production configuration and can trigger real API calls, database writes, or notifications.

***

## Connect to AI agents

{% tabs %}
{% tab title="Claude" %}
{% embed url="<https://youtu.be/byVl1GjQssw>" %}

You need a Nected account and a Claude plan that supports custom connectors. Add Nected once, and it works on the web, in Claude Desktop, in the mobile apps, and in Claude Code.

1. In a Claude chat, click **+**, then go to **Connectors → Add connector → Add custom connector**.

   <figure><img src="https://4290782554-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLg716fCfV8IUwXQygkTG%2Fuploads%2FDK4NmLFmpf8IJ6og1oGh%2Fimage.png?alt=media&#x26;token=88738dda-a0c6-43d7-a637-da351b3ad8eb" alt=""><figcaption></figcaption></figure>
2. Enter **Name** `Nected MCP` and **Remote MCP server URL** `https://api.nected.ai/mcp`. Keep the default authentication settings and click **Add**.

   <figure><img src="https://4290782554-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLg716fCfV8IUwXQygkTG%2Fuploads%2FDUEwqfUMAnongnX3fxxp%2Fimage.png?alt=media&#x26;token=db36fb17-2e80-4d2b-a5c1-3ea153bdbe72" alt=""><figcaption></figcaption></figure>
3. On Nected's connection page, sign in, select your workspace, and click **Continue**.

   <figure><img src="https://4290782554-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLg716fCfV8IUwXQygkTG%2Fuploads%2FMHh8QhCDYlcufGXL6viA%2FScreenshot%202026-10-28%20at%2012.36.35%E2%80%AFPM.png?alt=media&#x26;token=6659f061-4d10-4248-893c-aa144243afba" alt=""><figcaption></figcaption></figure>
4. In a chat, open the **Connectors** menu and turn on **Nected MCP**.

   <figure><img src="https://4290782554-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FLg716fCfV8IUwXQygkTG%2Fuploads%2FtlSpaiwDugn6yr5oaXOE%2Fimage.png?alt=media&#x26;token=b044399a-a13e-477e-914d-d3e95292d746" alt=""><figcaption></figcaption></figure>
5. Ask *"List the rules in my Nected workspace"* to check that it works.

{% hint style="info" %}
In **Claude Code**, Nected MCP is available automatically when you're signed in with the same Claude account. On **Team and Enterprise** plans, an Owner may need to add the connector first.
{% endhint %}

For more details, see Claude's guide: [Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp).
{% endtab %}

{% tab title="ChatGPT" %}

1. Open [ChatGPT Plugins](https://chatgpt.com/plugins).
2. Click **+**, then **Add custom MCP server**.
3. Enter a name, such as `Nected MCP`, and a short description.
4. Under **Connection**, enter the server URL `https://api.nected.ai/mcp`.
5. Configure authentication, review the warning, and select **I understand and want to continue**.
6. Select **Create as a plugin**, then review the Nected tools that ChatGPT discovers.
7. When prompted, sign in to Nected, select your workspace, and click **Continue**.

Custom MCP servers are subject to your ChatGPT account and workspace policies. For details, see OpenAI's guide: [Connecting custom MCP servers in ChatGPT](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt).
{% endtab %}

{% tab title="Grok" %}
Grok connects to remote MCP servers through the xAI API, as a tool in a Responses API request.

1. Get an xAI API key from the xAI console.
2. In your request, add Nected as a tool of type `mcp`:

   ```json
   "tools": [
     {
       "type": "mcp",
       "server_url": "https://api.nected.ai/mcp",
       "server_label": "nected",
       "authorization": "<your Nected access token>"
     }
   ]
   ```
3. Send your prompt. Grok calls the Nected tools it needs and returns the result.

The `authorization` value is sent as the `Authorization` header to Nected. For details, see xAI's guide: [Remote MCP Tools](https://docs.x.ai/docs/guides/tools/remote-mcp-tools).
{% endtab %}

{% tab title="Gemini CLI" %}

1. Add Nected as a remote HTTP server:

   ```bash
   gemini mcp add --transport http --scope user nected https://api.nected.ai/mcp
   ```
2. Start Gemini CLI and run `/mcp auth nected`.
3. Your browser opens Nected's connection page. Sign in, select your workspace, and click **Continue**.
4. Run `/mcp` to check that **nected** is connected and its tools are listed.

Sign-in needs a local browser, so it doesn't work in headless environments. For details, see Google's guide: [MCP servers with Gemini CLI](https://geminicli.com/docs/tools/mcp-server/).
{% endtab %}

{% tab title="Cursor" %}

1. Create or open `~/.cursor/mcp.json` to use Nected in every project, or `.cursor/mcp.json` to use it in one project.
2. Add Nected:

   ```json
   {
     "mcpServers": {
       "nected": {
         "url": "https://api.nected.ai/mcp"
       }
     }
   }
   ```
3. Open **Customize** in the sidebar and make sure **nected** is enabled.
4. When Cursor asks you to sign in, complete Nected's connection page: select your workspace and click **Continue**.

For details, see Cursor's guide: [Model Context Protocol (MCP)](https://cursor.com/docs/context/mcp).
{% endtab %}
{% endtabs %}

### Switch workspaces

Each connection works with one workspace. To use another one, disconnect Nected MCP in your AI agent, connect it again, and select the other workspace on Nected's connection page.

***

## Sample prompts

Use these prompts as a starting point. Replace the rule, workflow, and integration names with the ones in your workspace. Expand a prompt to see what the AI agent does after you send it.

### List entities

<details>

<summary>List rules by type and status</summary>

> *"List all decision tables in my workspace that are still in draft."*

**What happens**

1. The agent searches your rules, filtered to decision tables with the **draft** status.
2. It replies with a list showing each rule's name, type, status, and when it was last updated.

</details>

<details>

<summary>See which rules each workflow uses</summary>

> *"Show me the published workflows and the rules each one uses."*

**What happens**

1. The agent lists your published workflows.
2. It opens each workflow and reads its nodes.
3. It replies with each workflow and the rules it calls, including the rule version each node uses.

</details>

<details>

<summary>Find integrations and datasources</summary>

> *"Which integrations and datasources do we have for Postgres?"*

**What happens**

1. The agent lists your database integrations and picks out the PostgreSQL ones.
2. It lists the datasources built on each of them.
3. It replies with the integrations, their datasources, and the fields each datasource returns. Credentials stay masked.

</details>

### Update dependencies inside a rule or workflow

<details>

<summary>Pin a workflow step to a published rule version</summary>

> *"In the Loan Approval workflow, point the Credit Score step to published version 0.3.0 of the Credit Score decision table instead of the draft."*

**What happens**

1. The agent opens the latest draft of the Loan Approval workflow.
2. It finds the Credit Score step and changes the rule version it uses to **0.3.0**. Nothing else in the workflow changes.
3. It saves the workflow as a draft and confirms the change. The live version keeps working as before until you publish.

</details>

<details>

<summary>Replace and disable rules in a ruleset</summary>

> *"In the Pricing ruleset, replace Old Discount with Order Discount and disable Festive Offer."*

**What happens**

1. The agent opens the Pricing ruleset and finds the Order Discount rule in your workspace.
2. It swaps Old Discount for Order Discount in the same position and turns off Festive Offer.
3. It saves the ruleset as a draft and shows you the new order of rules.

</details>

<details>

<summary>Switch a rule to a different datasource</summary>

> *"Update the Risk Tier rule to read customer data from the customers\_v2 datasource instead of customers."*

**What happens**

1. The agent opens the Risk Tier rule and the customers\_v2 datasource.
2. It checks that customers\_v2 returns the fields the rule depends on, and tells you if any are missing.
3. It updates the rule to use customers\_v2 and saves it as a draft.

</details>

### Test

<details>

<summary>Test a rule with several inputs</summary>

> *"Test the Loan Eligibility rule with age 30 and credit score 720, then with age 19 and credit score 800, and show the results side by side."*

**What happens**

1. The agent runs a test of the Loan Eligibility draft for each set of inputs. Nothing is published.
2. It replies with a table of the inputs and outputs for each test, and explains why each result came out the way it did.
3. If the tests pass, the rule's status moves to **tested**.

</details>

<details>

<summary>Test each path of a workflow</summary>

> *"Test the Ticket Triage workflow with a high-priority and a low-priority issue, and tell me which steps ran in each case."*

**What happens**

1. The agent tests the workflow draft twice, once with each issue.
2. For each test, it shows the output and the steps that ran. For example, the on-call alert runs only for the high-priority issue.
3. If something fails, it points to the step that failed and suggests a fix.

</details>

### Audit

<details>

<summary>Find failed executions</summary>

> *"Show the last 10 failed executions in production."*

**What happens**

1. The agent searches the audit log for failed runs in the **production** environment, newest first.
2. It replies with a list showing each run's time, rule or workflow name, trigger (API, schedule, webhook, and so on), and duration.

</details>

<details>

<summary>Find out why a run failed</summary>

> *"Why did the last run of the KYC Onboarding workflow fail? Show me the step that errored and its input."*

**What happens**

1. The agent finds the most recent run of KYC Onboarding in the audit log.
2. It opens that execution and reads the result of each step.
3. It replies with the step that failed, the error message, and the input that step received, then suggests what to check.

</details>

<details>

<summary>Get usage numbers for a rule</summary>

> *"How many times was Loan Eligibility executed through the API yesterday, and how many of those failed?"*

**What happens**

1. The agent searches the audit log for Loan Eligibility runs triggered through the API during yesterday's date range.
2. It counts the successful and failed runs.
3. It replies with the totals and, if there were failures, what they had in common.

</details>

***

## Troubleshooting

<details>

<summary>The agent doesn't use Nected</summary>

Make sure the Nected MCP connector is turned on for that chat or session.

</details>

<details>

<summary>Nected MCP is missing in Claude Code</summary>

Sign in to Claude Code with the same Claude account where you added the connector, then run `/mcp`.

</details>

<details>

<summary>Nected's connection page doesn't open</summary>

Allow pop-ups for your AI agent's site, then connect again.

</details>

<details>

<summary>"Unauthorized" or session-expired errors</summary>

Disconnect and reconnect Nected MCP.

</details>

<details>

<summary>The agent can read rules but can't change or publish them</summary>

Your Nected role doesn't allow it. Ask a workspace admin.

</details>

<details>

<summary>You see another workspace's data</summary>

You connected a different workspace. See Switch workspaces.

</details>

<details>

<summary>The option to add a custom connector is missing</summary>

Your plan doesn't support custom MCP connectors, or an admin must add the connector for your organization first.

</details>

<details>

<summary>Nected tools don't appear after connecting</summary>

Start a new chat or session so the agent loads the Nected tools. If they still don't appear, disconnect Nected MCP and connect it again.

</details>

<details>

<summary>The connection worked before but has stopped</summary>

The connection may have been removed, or your access to the workspace may have changed. Reconnect Nected MCP and select the workspace again.

</details>

<details>

<summary>Nected's connection page shows an error or has expired</summary>

Close the page and start the connection again from your AI agent.

</details>
