AI agents that work through your MCP servers, under your own rules. } subtitle="Connect your MCP servers and your own code once, then choose which agents use them. Web research and code execution come built in. Each Task returns one result your team can use." primaryCta={{ label: "Plan your first use case with us", dialog: "contact" }} secondaryCta={{ label: "Start with the Python SDK", href: "https://duale.ai/en/docs/sdk.md" }} />

Connect the servers your teams and vendors already run, then give their Tools to the agents that need them. Turn the Python functions in your application into Tools. Every change they make follows your business rules. Every Task can search the public web, read the pages it needs and cite them in its result. Every Task can run Python or JavaScript to calculate, compare and transform data. Each run is isolated from your systems.

Connect each MCP server once, with its own credentials. Every agent you assign uses that connection, so no application team rebuilds the integration. Give each agent only the systems its work needs. Your support agent reaches the service desk, and your finance agent reaches the ERP. In the Dashboard, allow or block an MCP Connection's calls, or remove an agent from it. Every Task picks up the change at its next step. Each agent uses its Tools under [its own identity](https://duale.ai/en/product/identity-and-access.md). [Usage analytics](https://duale.ai/en/product/usage-analytics.md) shows the success rate of each MCP Connection. For your company documents, agents use [Libraries](https://duale.ai/en/product/libraries.md).

Your ERP, databases and internal tools stay private. Your Python functions run in your own application, inside your network, and your application opens the connection to Duale AI. The functions read and update those systems with your own permissions. When a change needs a person's decision, your function can record it as a proposal for that person to approve. ```mermaid flowchart LR accTitle: How an agent reaches your private systems accDescr { Inside your network, your Python functions apply your rules to your private systems and open an outbound connection. Your MCP servers, Web research and code execution complete the Tools you gave the agent. Each delegated Task uses those Tools and returns one result to your team. } Systems["Your private systems"] Functions["Your Python functions, inside your network"] MCP["Your MCP servers"] BuiltIn["Web research and code execution"] Tools["The Tools you gave this agent"] Task["Delegated Task"] Result["One result for your team"] Functions -->|apply your rules| Systems Functions -->|outbound connection| Tools MCP --> Tools BuiltIn --> Tools Tools --> Task Task --> Result ```

In one Task, the agent reads the open tickets in your service desk through its MCP server. Your own function returns the usage and invoices from your ERP. Web research finds the customer's latest public news, and code execution calculates the quarter's trends. The account manager receives one review: - **Usage**: up 14% on the previous quarter. - **Support**: three open tickets, all about delivery times; the oldest is 12 days old. - **News**: the customer announced a new distribution site on 2 October, with a link to its press release. - **Next step**: a volume tier at renewal, recorded by your function as a proposal for the account manager to approve.

Get a sourced brief on each account before the meeting, with the new facts recorded in your CRM. Your own code makes every write to your records. Get one due diligence memo with every finding sourced. Web research covers the press, and the figures are calculated from the accounts.

A standard Task costs €0.10 and a fast Task €0.20, excluding VAT. [See pricing](https://duale.ai/en/product/pricing.md) Use the managed service, hosted in the European Union, or run the Platform in your own infrastructure. Add agents as your teams and use cases grow.

## Questions about Tools ### Can an agent's Tools use a test copy before our live systems Yes. The agent can work on a test copy, such as your CRM sandbox, before you grant it the live system. [See which systems your agents can count on in Usage analytics](https://duale.ai/en/product/usage-analytics.md), then [see when to create separate connections](https://duale.ai/en/docs/tools/mcp-connections.md#before-you-create-the-connection). ### Do our MCP servers keep their own access rules when our agents call them Yes. The permissions you set in your own system keep applying, whichever agent calls your MCP server. [See how Identity and access limits each agent](https://duale.ai/en/product/identity-and-access.md), then [see which account reaches each server](https://duale.ai/en/docs/tools/mcp-connections.md#know-which-identity-reaches-the-server). ### Can we see what an agent did in our systems Yes. In the Dashboard, each Task shows every call it made to your Tools and how each one ended. [See how Usage analytics follows each Task](https://duale.ai/en/product/usage-analytics.md), then [read what each outcome means](https://duale.ai/en/docs/agent-harness.md#read-a-tool-call-outcome). ### Does a Task cost more when it calls more Tools No. A Task keeps its fixed price, standard or fast, Web research and code execution included. [See the price per Task](https://duale.ai/en/product/pricing.md), then [see the Tools built into every Task](https://duale.ai/en/docs/tools.md#built-in-tools). ## Related content - [Create and manage MCP Connections](https://duale.ai/en/docs/tools/mcp-connections.md) - [Tool limits: Source, MCP catalog, and Agent Identity](https://duale.ai/en/docs/tools/limits.md) - [Tools: sources, Built-in Tools, and source choice](https://duale.ai/en/docs/tools.md) - [Documentation: integration, capabilities, and limits](https://duale.ai/en/docs.md) - [Connect an MCP client to the documentation server](https://duale.ai/en/docs/mcp.md) - [MCP server compatibility, authentication, and call contract](https://duale.ai/en/docs/tools/reference.md) --- ## Sitemap See the full [Markdown sitemap](https://duale.ai/sitemap.md) for all pages.