AI integration guides

Why integration matters

Model Context Protocol (MCP) provides a standardized way to connect agents to external systems — databases, APIs, file services, and enterprise tools — enabling workflows that combine AI reasoning with real-world data and actions.

For Indian deployments, determine whether personal data is processed and review the current DPDP Act, rules, government notifications, contracts, and security controls with qualified counsel. This page does not infer a blanket India-only data-residency rule or treat consent as the sole basis for every cross-border transfer.

These integration outlines are review checklists, not tested copy-paste procedures. They remain noindex until each client, server, configuration path, and security step is checked against current first-party documentation.

Before you start any of these guides

Every guide below assumes an MCP-compatible client is already installed and running — the steps cover connecting that client to a specific server, not installing the client itself. If a step references a config file path (like ~/.cline_mcp_settings.json), confirm the current path against that client's own documentation first: client config locations change across versions more often than the underlying MCP connection steps do.

Read-only first, write access second

Where a guide below touches a database, a payment system, or a code repository, start with the most restrictive access mode available — read-only database credentials, a scoped-down GitHub token, sandbox-mode payment keys — before granting write access. An agent that can only read cannot accidentally execute a destructive query, merge an unreviewed pull request, or trigger a real payment; expand scope only after you have watched the agent operate correctly under the restrictive mode.

Common failure points

Most first-time MCP integration failures fall into a small number of categories: a transport mismatch (client configured for stdio while the server expects Streamable HTTP, or vice versa), a credential with the wrong scope (a GitHub token missing the repo scope, a database user without SELECT on the target schema), or a firewall/network boundary blocking a remote server the client expects to reach directly. When a connection fails, check these three before assuming the server itself is broken — most MCP servers log a clear error for exactly this class of misconfiguration.