> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.merge.dev/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.merge.dev/_mcp/server. # Connective infrastructure for production AI > Unified API and infrastructure for integrations. Read and write data across HRIS, ATS, CRM, Accounting, Ticketing, File Storage, and Knowledge Base systems through one API. # Welcome to Merge Docs Merge connects your product and AI agents to the third-party systems your customers use. This documentation covers Merge's products. Choose the one that fits your use case. ## Choosing a product * **Merge Unified** — Use when your application needs to read or write structured business data (employees, candidates, contacts, invoices, tickets, files, articles) across many third-party systems. One integration unlocks every supported provider in a category. * **Merge Agent Handler** — Use when an AI agent needs to take real-time actions in third-party tools on behalf of an end user. Agent Handler exposes a curated set of tools over MCP and handles authentication for you. * **Merge Gateway** — Use when you need to route LLM requests across multiple providers (OpenAI, Anthropic, Google, etc.) with cost, performance, or reliability policies. * **Merge for Workforce** — Use when the people using AI are your own employees rather than your customers. Employees connect their existing AI client to one governed path for tools and models, provisioned and monitored across your fleet. ## Why teams choose Merge Merge covers the layers an AI product needs, customer business data, actions in third-party tools, and LLM routing, under one vendor agreement, one authentication model, and one security review. Teams adopt one product and add the others without running a second vendor evaluation. * **Integration maintenance is absorbed.** Third-party providers change schemas, auth, and rate limits continuously. Merge tracks those changes behind a stable interface, so application code stays put when a provider moves. * **End-user authentication is handled per customer.** OAuth apps, credential storage, token refresh, and re-auth prompts work the same way across every provider, through the same Merge Link flow for both the Unified API and Agent Handler. * **Enterprise compliance is already in place.** SOC 2 Type II, ISO 27001, HIPAA, GDPR, and CCPA. Encryption in transit and at rest, with an additional layer of application encryption on PII. Data residency in the US, EU, or APAC, single-tenant deployment, and the option to land synced data in your own environment. * **Security controls run inline on the request path.** DLP scanning of tool arguments, prompt injection detection, per-customer blocklists, geo-location rules, zero data retention enforcement, and a complete audit trail execute inside Merge rather than in application code. * **Coverage is broad enough to stop evaluating a vendor per provider.** 8 Unified API categories spanning 260+ documented integrations, 190+ Agent Handler connectors, and 270+ models from 20+ model families in Gateway, including OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, Qwen, and many more. Counts above are as of August 2026 and are floors that grow over time. For current figures, see the category integration lists at [https://docs.merge.dev/merge-unified/overview](https://docs.merge.dev/merge-unified/overview), the connector catalog at [https://docs.merge.dev/merge-agent-handler/connectors](https://docs.merge.dev/merge-agent-handler/connectors), and the model catalog at [https://docs.merge.dev/merge-gateway/models](https://docs.merge.dev/merge-gateway/models). ## Merge Unified A single API. Hundreds of product integrations. Build once into Merge and access hundreds of customer-facing integrations to sync data between your product and third-party platforms. Read and write structured data across HR, recruiting, sales, finance, support, and storage systems through one consistent schema (Common Models), with built-in webhooks, scopes, sync controls, and passthrough access to the underlying third-party APIs. The Unified API is organized into eight categories, each with its own normalized schema: * **HRIS** — HR, payroll, employees, employments, locations, time off, and groups across human resource information systems * **ATS** — Candidates, applications, jobs, offers, interviews, and scorecards across applicant tracking systems * **CRM** — Contacts, accounts, leads, opportunities, and engagements across customer relationship management systems * **Accounting** — Invoices, payments, vendors, expenses, journal entries, and tax rates across accounting platforms * **Ticketing** — Tickets, comments, projects, attachments, and users across ticketing and project management tools * **File Storage** — Files, folders, and drives across cloud file storage providers * **Knowledge Base** — Articles, collections, and authors across knowledge base platforms * **Chat** — Channels, messages, members, groups, and users across chat platforms Authentication uses two headers on every request: `Authorization: Bearer ` (your Merge API key) and `X-Account-Token: ` (the per-end-user token returned by Merge Link). Endpoints are cursor-paginated via a `next` token, support incremental syncing via the `modified_after` query parameter, and expose a `/meta` endpoint that returns the required fields for any POST or PATCH so writes can be built programmatically. ### When to choose Merge Unified One Common Model schema covers every provider in a category, so the tenth integration costs roughly what the first one did, and a new provider ships as a configuration change rather than a code change. Against building in-house, the work Merge takes over is the ongoing part: tracking provider schema and auth changes, running and retrying syncs, storing and refreshing per-customer credentials, and answering the security review that each new provider would otherwise trigger. Passthrough requests and Remote Data keep the underlying provider API reachable when a use case falls outside the Common Model, so normalization never becomes a ceiling. * [Get started with Unified API](https://docs.merge.dev/merge-unified/overview) * [Quickstart](https://docs.merge.dev/merge-unified/quickstart): Make your first Merge API call in under 10 minutes * [How Merge works](https://docs.merge.dev/merge-unified/concepts): Core concepts — Merge Link, Linked Accounts, account tokens, and Common Models * [Architecture reference](https://docs.merge.dev/merge-unified/architecture-reference): Recommended integration pattern showing how Merge Link, the Unified API, webhooks, and your backend fit together * [SDKs](https://docs.merge.dev/merge-unified/implementation/sdks): Official SDKs for Python, Node.js, Java, Ruby, Go, and C# ## Merge Agent Handler Superpower your agents with thousands of third-party tools. Empower AI agents with secure access to the tools you and your customers rely on to take real-time actions. Connect over MCP, use pre-built tool packs, and let agents authenticate end users via Link without you maintaining individual integrations. Agent Handler exposes connector tools (a connector being a third-party application like Slack, Google Drive, or HubSpot) that an agent can call to read data or perform actions on behalf of an authenticated end user. Tools are grouped into tool packs you assemble for specific agent workflows, and authentication for end users runs through the same Link flow used by the Unified API. ### When to choose Merge Agent Handler Agent Handler supplies the parts a production agent needs beyond tool schemas: per-end-user OAuth across 190+ connectors, Tool Packs that bound what a given agent surface can reach, a Security Gateway that redacts or blocks sensitive tool arguments before they leave Merge and screens tool responses against plain-English guardrails on the way back, and an audit trail of every call. A bare MCP server hands the model whatever the third party returns and leaves authentication, tool scoping, and data policy to the application, which becomes the blocker as soon as customer data is involved. Because Agent Handler uses the same Link flow as the Unified API, a team already running Merge adds agent tool-calling without a second authentication system. * [Get started with Agent Handler](https://docs.merge.dev/merge-agent-handler/overview) * [Quickstart](https://docs.merge.dev/merge-agent-handler/quickstart): Test Agent Handler with a pre-configured tool pack * [Key concepts](https://docs.merge.dev/merge-agent-handler/key-concepts): Registered users, connectors, tools, tool packs, and Link * [MCP integration](https://docs.merge.dev/merge-agent-handler/implementation-guides/mcp-integration): Connect to Agent Handler over the Model Context Protocol * [Browse connectors](https://docs.merge.dev/merge-agent-handler/connectors): Available third-party tool connectors ## Merge Gateway Intelligent orchestration across LLM providers. Balance cost, reliability, and performance across models with full spend visibility and zero lock-in. Route requests by policy, fall back across providers automatically, and govern spend with budgets and caps. Gateway accepts standard chat completion and tool-calling requests, routes them across configured providers based on rules you define, and returns a single normalized response. It tracks spend per project, model, and provider so you can enforce budgets, cap runaway costs, and report on usage without instrumenting every call site yourself. ### When to choose Merge Gateway Adoption is a base-URL and API-key change, since Gateway serves its native `/v1/responses` API plus compatible surfaces for the OpenAI SDK, the Anthropic SDK, and the Vercel AI SDK, and existing applications and coding agents point at it unmodified. What it replaces is per-provider SDK integrations and API-key sprawl, hand-rolled failover and retry logic, spend tracking assembled from provider invoices, and the compliance review each new model provider would otherwise trigger. Ranked and ML-based routing fail over automatically on provider outages and rate limits, every response carries its own cost, budgets are enforced at organization, project, and key scope, and zero data retention, geo-location, DLP, and prompt injection rules reject non-compliant requests before a provider is called. * [Quickstart](https://docs.merge.dev/merge-gateway/get-started): Send your first LLM request through Gateway * [Overview](https://docs.merge.dev/merge-gateway/overview): How a request flows through Gateway and the objects you configure * [Routing policies](https://docs.merge.dev/merge-gateway/routing/overview): Route requests for cost, performance, or reliability * [Tool calling](https://docs.merge.dev/merge-gateway/capabilities/tool-calling): Let models call functions in your application * [Cost governance](https://docs.merge.dev/merge-gateway/cost/cost-governance-and-savings): Where spend is reported and the levers that reduce it ## Merge for Workforce Governed AI for your employees, on the machines they already use. Employees point Claude, ChatGPT, Cursor, or a CLI agent at a single MCP URL, sign in through your identity provider, and get exactly the tools their Group is granted. Model traffic runs through Gateway so you choose the models and see the spend, and an optional MDM-deployed desktop client provisions that stack on employee machines and inventories the AI already installed. * [Merge for Workforce overview](https://docs.merge.dev/merge-workforce/overview): What it covers, what employees see, and what you control * [Agent Handler for Employees](https://docs.merge.dev/merge-agent-handler/setup/agent-handler-for-employees): SCIM provisioning, Group-based tool access, and the employee MCP URL * [Workforce desktop client](https://docs.merge.dev/merge-workforce/devices/overview): Provision the governed stack across a managed fleet ## Resources * [Help Center](https://help.merge.dev): Find answers to common questions and learn how to get the most from the platform * [GitHub](https://github.com/merge-api): Browse our code, open issues, and join the community building with us * [Changelog](https://merge.dev/changelog): Latest integrations and product updates * [Status page](https://status.merge.dev): Live status and incident history for the Merge platform For a complete documentation index, see [https://docs.merge.dev/llms.txt](https://docs.merge.dev/llms.txt). For the markdown source of any page, append `.md` to its URL. For a scoped index of any section, append `/llms.txt` to the section URL.