10 Merge.dev Alternatives for Provider-Specific Fields in 2026
Merge's Common Model can drop provider-specific fields your agent needs. Ten alternatives compared on how they reach those fields, with prices checked November 2026, and where Swytchcode governs the direct call.
Key takeaways
- -Teams leave Merge when the field an agent needs is not in the Common Model.
- -Merge's Remote Data and Authenticated Passthrough are on Professional and Enterprise, and Remote Data only refreshes when a Common Model field changes.
- -Apideck, Unified.to, Kombo, and Finch are other unified APIs. Each reaches unmapped fields in a different way.
- -Nango, Paragon, and Pipedream let you write or proxy the provider call yourself.
- -A passthrough call skips the unified guardrails, so policy, safe retries, and audit become your job.
- -Swytchcode governs that direct call: brokered OAuth, policy-as-code, idempotent retries, and a local audit log, on infrastructure you own.
Merge.dev is a unified API. You build against one Common Model for ATS, HRIS, CRM, accounting, ticketing, and file storage, and Merge maps each provider into it. Teams look for an alternative when the field their agent needs is not in that model: a custom field a customer created, a status only one provider has, or an object Merge never mapped. Then they want to reach the raw provider data without a contract upgrade, own the integration code, or make sure the call they make outside Merge is governed and retried safely.
The ten options below are grouped by that job. They are not a ranked scoreboard. Apideck, Unified.to, Kombo, and Finch are other unified APIs, each with its own way to reach fields outside the common schema. Nango, Paragon, and Pipedream let you write or proxy the provider call yourself. Composio and Arcade are tool catalogs for agents. Swytchcode is a different layer: it executes the API call, including the provider-specific call your unified model does not cover.
Quick verdict: Keep Merge if the Common Model covers the screen you sell and the missing field is rare. Field Mappings or Remote Data may close the gap on a Professional contract. Move to a pass-through unified API such as Unified.to or Apideck if you need live provider data, or to Nango if you want the integration code in your repo. Whatever you pick, the call that bypasses the unified model still needs auth, policy, safe retries, and an audit log. That call is what Swytchcode governs.
Why teams look for a Merge.dev alternative
Merge is a good fit when every customer runs a different ATS or HRIS and your product needs the same Candidate, Employee, or Invoice shape from all of them. The friction starts when one field does not fit that shape. Merge's own docs list several ways around it. Each one has a catch when the consumer is an agent.
The field exists in the provider, not in the Common Model
A Common Model keeps the fields most providers share. A field one provider invented, or one a customer added, has no slot. Merge lists Field Mappings to add or override Common Model fields, Remote Field Classes to read and write non-mapped data in a standard format, Native Enum Values to see a provider's own status values, and Custom Data for custom objects. These are real tools. They also mean the field your agent needs is now configured per provider and per customer, outside the schema you integrated against. Source: Merge supplemental data overview.
Remote Data is a snapshot, not a live read
Remote Data returns the provider's original payload next to the normalized record when you add include_remote_data=true. Merge's docs say it is only available on Professional and Enterprise, and that Remote Data is only updated in a sync if a Common Model field changes. So a provider-only field can change in the source system and stay stale in Merge until something else on the record moves. An agent answering a recruiter in real time can read an old value and not know it. Source: Merge Remote Data docs.
Passthrough is a raw call with none of the unified guardrails
Authenticated Passthrough sends a request straight to the provider with the Linked Account's stored credentials, in the provider's own format. It is also Professional or Enterprise only, per Merge's passthrough docs. Once you use it, the agent is making a raw provider call. The unified model's validation, error shape, and pagination no longer apply. Which methods are allowed, whether a retried POST creates a second record, and what gets logged are now your problem. That is the class of failure described in Why your AI agent breaks in production.
The fix sits on a contract plan
Merge publishes its Launch price: 3 production Linked Accounts free, then $650 a month for up to 10, then $65 per Linked Account. Launch syncs daily. Custom fields, field-level scopes, and custom sync frequency are listed on Professional, which is contract priced. A team that only needs one provider field ends up negotiating a contract to get it. Source: Merge pricing.
How to evaluate a Merge.dev alternative
Use these criteria to compare any Merge alternative, including the ones below. For this job, read catalog coverage as field coverage: can you reach the provider-specific field, and how. Each row is something you can check in the vendor's docs or a trial before you commit.
| Criterion | What to verify |
|---|---|
| Identity and auth | Whose credentials each call runs with: one connection per customer or each end user's own OAuth grant. Where tokens are stored, and whether the model ever sees them. |
| Catalog coverage | Whether the providers you need are covered, and whether the provider-specific field is reachable through the unified model, a mapping, raw data, or a passthrough call. |
| Write safety | Whether retries are automatic, and whether a retried POST through the unified API or a passthrough call can create a duplicate record. |
| Policy and approval | Whether you can allowlist actions, block calls by argument, and hold risky calls for a person, including calls that bypass the unified model. |
| Error handling | Whether failures come back in one consistent shape, including passthrough calls that return the provider's raw error. |
| Audit | What is logged for each unified and passthrough call, where the log lives, and how long it is kept. |
| Hosting | Vendor cloud, your own cloud, or self-hosted, and whether customer data is stored by the vendor or passed through. |
| Pricing model | What is metered (linked accounts, consumers, connections, API calls, or tool calls) and which plan unlocks custom fields and passthrough. |
How the ten alternatives compare
Prices are what each vendor published on its own site in November 2026, or quote only when the pricing page has no list price. Closest job describes how each one handles a provider field the common schema does not carry.
| Alternative | Closest job | How the agent reaches it | Public starting price |
|---|---|---|---|
| Apideck | Unified API with a proxy for unmapped endpoints | Unified REST API, Proxy API | Launch $599/mo for 25 consumers |
| Unified.to | Real-time pass-through unified API | Unified REST API, Passthrough, MCP | Test free (non-production); Grow from $750/mo |
| Kombo | HR and ATS unified API with remote fields | Unified API, Passthrough API | Quote only (platform fee plus per-customer fee) |
| Finch | Employment and payroll data | Unified API, Request Forwarding | Starter $65 per connection/mo |
| Nango | Integrations you own as code | TypeScript functions, proxy, API, MCP | Free, 10 connections; pay-as-you-go $50/mo |
| Paragon | Embedded integrations in your product | ActionKit API or MCP, Managed Sync | Quote only |
| Pipedream | Broad catalog with a per-user proxy | Remote MCP, SDK, or Connect proxy | Free to build; Startup $99/mo annual |
| Composio | Hosted tool catalog for agents | MCP and SDKs | Free, 100K tool calls/mo; Pro $29/mo |
| Arcade | Per-user tool calls | Python and JavaScript SDKs, MCP | Free, 2,000 tool calls/mo |
| Swytchcode | Governed execution of each API call | CLI, MCP, or Runtime SDKs | Free, 100 live commands/mo |
1. Apideck
Best for: SaaS products that want a unified API with a proxy for endpoints the unified model does not cover yet.
Apideck is the closest like-for-like swap. It covers accounting, CRM, HRIS, ATS, file storage, and more behind one API, and every plan includes the Proxy API, which Apideck describes as the way to reach endpoints not yet supported on a Unified API. Raw mode adds the provider's original response as _raw next to the normalized data, on operations that document the raw parameter. Apideck's own raw mode guide says raw data does not make provider fields portable. For that you need custom field mapping.
The catch for this job is the plan line. On the November 2026 pricing page, Launch is $599 a month for 25 consumers and one API category. Custom field mapping and data scopes start on Scale, at $1,299 a month for 100 consumers. Choose Apideck when the provider-specific field is a custom field you can map per customer, or an endpoint you are happy to proxy.
2. Unified.to
Best for: Products that need live provider data on every read, across many categories, without a stored copy.
Unified.to sells a real-time, pass-through unified API. Requests go to the provider when you make them, instead of reading from a synced store, which is the opposite trade to Merge's Remote Data. Its own comparison post claims 2,038 integrations across 38 categories on every plan. Treat that as Unified.to's figure. Its pricing page lists the Passthrough API on Test, Grow, Pro, and Scale.
Pricing is metered on API calls. Test is free for 30 days and non-production only. Grow starts at $750 a month with 750,000 calls, then $1.00 per 1,000. Pro starts at $1,500 a month with 2 million. Pick Unified.to when stale synced data is the real problem. An agent that loops will spend API calls, so put a ceiling on it.
3. Kombo
Best for: HR, ATS, payroll, and LMS products that need custom fields on reads and tool-specific data on writes.
Kombo's docs list four ways past the unified model. Custom fields map a key you define to each integration's field and return it in a custom_fields object. Remote fields let you send tool-specific data on a write, following each tool's own schema. Field remapping changes which source field fills a unified field. The Passthrough API calls the tool's native API with Kombo handling authentication. Custom fields read from synced data, with a Custom Field Explorer for live lookups. Source: Kombo, extending the model.
Kombo does not publish list prices. Its pricing page describes a fixed annual platform fee plus a per-customer fee that drops as you scale, and lists custom fields and field remapping on every plan. Choose Kombo when the missing field is an HR or ATS custom field and you want the mapping flow handled for you, including self-serve mapping by your customers.
4. Finch
Best for: Benefits, payroll, and HR products that need employment data from many payroll providers.
Finch is a unified API for employment systems: organization, payroll, deductions, and documents. The November 2026 pricing page lists Starter at $65 per connection a month, up to 15 connections and 24 providers, read-only, for employers under 500 employees. Pro and Premier are quoted, with all 250+ providers. For fields outside the model, Pro and Premier include Request Forwarding, which makes a direct provider call and returns the raw response.
Choose Finch when the job is payroll and deductions, not a general ATS or CRM. Request Forwarding is a raw call, so the same write-safety questions apply as with Merge's Passthrough: who allows the method, and what happens on a retry.
5. Nango
Best for: Teams that want each integration, including provider-specific fields, written as code in their own repository.
Nango drops the fixed common model. You write each sync or action as a TypeScript function, deploy it, and expose it through Nango's API or MCP, with managed OAuth underneath. If a provider has a field nobody else has, you just read it. A Request Proxy covers one-off calls. That is more work than Merge on day one. It is less work on the day the field you need is not in anyone's schema.
The November 2026 pricing page lists Free at 10 connections, 10 compute hours, and 10 GB of transfer, hard capped. Pay-as-you-go is $50 a month including $50 in credits, at $0.29 per connection and $0.72 per compute hour. Enterprise adds self-hosting and bring-your-own-cloud. How it differs from call execution: Swytchcode vs Nango.
6. Paragon
Best for: SaaS products that embed integrations for their customers and want synced data plus real-time actions.
Paragon has three products: Managed Sync for ingestion, ActionKit for synchronous actions, and Workflows. ActionKit exposes hundreds of CRUD actions across dozens of integrations through one API and an MCP server, with agent-optimized schemas. Managed Sync can run as often as every minute, and its normalized schemas can be customized. Dynamic field mapping is listed on Enterprise.
Paragon prices by connected users, meaning the customer organizations using your integrations. No list price was on the pricing page reviewed in November 2026. Choose Paragon when the agent is a feature inside software you sell and each customer has to connect their own tools.
7. Pipedream
Best for: Agents that need a broad app catalog and a proxy that attaches each user's token to a raw call.
Pipedream Connect stores each external user's credentials and gives agents prebuilt actions and a remote MCP server. When a prebuilt action does not expose the field you need, the Connect proxy sends the HTTP request you specify and attaches that user's token. That is the raw provider call, with no unified schema in the way. The proxy does not tell the agent which fields or versions the provider actually accepts.
Free covers development. Production needs Startup: $99 a month billed annually or $150 monthly, with 100 external users and 10,000 usage credits. Extra users are $2 a month and extra credits $0.012. A proxy request uses credits like any action. Details: Swytchcode vs Pipedream.
8. Composio
Best for: Internal agents that need a wide hosted tool catalog quickly.
Composio is not a unified API. It is a catalog of prebuilt tools for agents with managed OAuth, and its pricing page lists 1,500+ toolkits on every plan. Each toolkit is shaped around one provider's API, so provider-specific fields are often present where a common schema would flatten them. You use the tools as shipped.
Hobby is free with 100K tool calls a month, up to 20K on Composio-managed apps. Pro is $29 a month with $29 in credits, then $0.0003 per call. Choose Composio when you want breadth fast for an internal agent. Comparison: Swytchcode vs Composio.
9. Arcade
Best for: User-facing agents that must act with each person's own OAuth grant.
Arcade is an actions runtime built around per-user authorization. If a user has not granted a scope, Arcade runs the OAuth flow. Its tools are written per provider, so like Composio it avoids a cross-provider schema. It does not replace a unified HRIS or ATS model for a product that renders the same table for every customer.
Free is $0 with 2,000 tool calls a month. Team is a $25 monthly platform fee plus $0.01 per tool call. Comparison: Swytchcode vs Arcade.
10. Swytchcode
Best for: Production agents that call provider APIs directly, including the fields a unified model drops, and need policy, approval, safe retries, and an audit trail on every call.
Swytchcode is last because its focus is the execution layer under the agent, not the unified model. It does not normalize ATS records into one schema. It runs the provider call itself: 360+ integrations in the catalog, plus any OpenAPI spec you bring for an internal or legacy API. That is the call you end up making when Merge's Common Model does not carry the field.
For each call, Swytchcode resolves the right HTTP method, validates the input, applies allow and deny rules from tooling.json and policies.json, injects OAuth, retries transient failures with backoff that honors Retry-After, and logs the call and the decision locally for 90 days with sensitive values redacted. POST and PATCH are retried only with an idempotency key, so a retry never creates a second record or charge. Requests go straight from your machine to the provider. There is no proxy hop.
Free is $0 forever for 100 live commands a month with allow and deny rules. Pro is $29 a month or $290 a year for 10,000. Business is $149 a month or $1,490 a year for a million, and adds approval workflows, team seats, and bring-your-own providers. Enterprise is custom and runs in your own cloud. It works as a CLI or an MCP server. The short version of the idea is What is an execution layer?. How a retry stays safe is in How Swytchcode retries failed API calls and Idempotency explained.
Why Swytchcode is the production runtime for reliable AI agents
Swytchcode is an execution kernel that sits between your AI agents and the APIs they call. It runs on infrastructure you own, checks policy before every call, brokers OAuth, retries safely without duplicate writes, and logs every call and decision locally. The same kernel serves LangGraph, the OpenAI Agents SDK, the Anthropic SDK, and MCP clients, across 75,000+ endpoints in 300+ services or your own OpenAPI spec, including legacy APIs.
Merge gives you one schema across providers. Swytchcode is the execution kernel that makes each production call, including the ones Merge's schema doesn't cover, governed and auditable, on your infrastructure.
Demos work. Production is a different animal.
Frameworks solve reasoning. None of them solve the moment an agent touches a real production API. That moment needs OAuth and credentials, permission control, policy enforcement, audit and compliance, and retries with idempotency. So every team builds the same middleware, then rebuilds it for the next framework.
| Production requirement | Build it yourself | With Swytchcode |
|---|---|---|
| OAuth and credentials | Token storage and refresh per framework | Brokered OAuth; tokens cached locally with AES-256-GCM, key in the OS keychain |
| Permissions and policy | Custom checks inside agent code | tooling.json allowlist plus policies.json rules, version-controlled, checked before every call |
| Human approval | Ad hoc scripts and chat bots | Matching calls held for approval in Slack; no answer in 48 hours means no run (Business and up) |
| Audit and compliance | Logs scattered across services | Every call and decision logged locally; cloud sync is opt-in |
| Retries and idempotency | Hand-written per API | Backoff that honors Retry-After; POST and PATCH retried only with an idempotency key |
| Legacy and internal APIs | A hand-written client per API | Bring your own OpenAPI spec; inputs validated before the call leaves your machine |
| Switching frameworks | Rebuild all of the above | Same kernel, same policy |
One execution kernel. Every agent framework.
- CLI execution kernel: runs on your infrastructure and decides what can run, and whether it should.
- Runtime SDKs: thin JavaScript and Python wrappers, so every framework hits the same kernel.
- Backend control plane: brokers OAuth, hosts the catalog, and aggregates telemetry. It never sees your payloads.
Governance your security team already recognizes
- Credentials never leave your infrastructure. OAuth tokens are cached locally with AES-256-GCM, and the key lives in the OS keychain.
- Policy-as-code. policies.json is version-controlled, so guardrails get reviewed like any other code. Allow and deny rules are on every plan, including Free.
- Fail-closed, with a human in the loop. Calls that match an approval policy are held and sent to your workspace's Slack or Telegram channel. If nobody approves within 48 hours, they never run. Approval workflows are on Business and Enterprise.
- Full local audit trail. Every outbound call and policy decision is recorded on your machine for 90 days, with sensitive values redacted. Cloud Sync is off by default and uploads summaries only, never payloads or credentials.
- No proxy hop. Requests go straight from your machine to the provider; Swytchcode does not proxy or store your API traffic.
- Deploy in your own cloud. Enterprise runs in your AWS, Azure, or GCP account, in the region you choose, with company SSO, role-based access, audit export, and an SLA.
- Coverage built in. 75,000+ endpoints across 300+ services, or your own OpenAPI spec for internal and legacy APIs.
Why enterprise teams choose Swytchcode
- It clears security review: execution and credentials stay inside your infrastructure.
- No framework lock-in: swap LangGraph for the OpenAI Agents SDK without rewriting policy.
- Reviewable by default: tooling.json and policies.json ship through normal code review.
Build it yourself and you own auth, retries, policy, and audit for every framework, forever. Swytchcode gives every agent the same governed path to production, so developers ship agents and platform teams keep control.
What ships today and what is next
Human approval ships today on Business and Enterprise: matching calls wait for a yes or no in Slack. Every integration's specs and documentation are observed for breaking changes and updates, which is what keeps an API agent-ready after the week you added it. Next on the roadmap are more policy stages, covering post-execution and streaming, and a Go runtime SDK built on the same thin-wrapper model. Pricing: Free is $0 forever for 100 live commands a month, with allow and deny rules. Pro is $29 a month for 10,000 live commands. Business is $149 a month for a million, and adds approval workflows, team seats and your own providers. Enterprise is custom and deploys in your own cloud account.
How do you make AI agents reliable in production?
Put one governed execution layer between every agent and every API. It should resolve credentials outside the prompt, enforce policy before the call, require approval for risky actions, retry without duplicate writes, return errors in one consistent shape, and keep an audit log. Swytchcode does this as a single kernel for every framework. Start with the MCP quickstart or read how the execution pipeline works.
Which alternative to pick
| If the requirement is | Start with | Leave it when |
|---|---|---|
| The common schema covers the field, or a mapping can add it | Merge, Apideck, or Kombo | The field changes live and a synced copy goes stale |
| Live provider data on every read | Unified.to | API-call metering does not fit an agent that loops |
| The field is a payroll or deduction detail | Finch | You need ATS or CRM data too |
| The integration code must live in your repo | Nango | You only need a prebuilt action today |
| Each customer connects their own tools inside your product | Paragon or Pipedream | You need one normalized table across every customer |
| The agent calls the provider directly and that call must be governed | Swytchcode | You need one normalized schema across many customers' systems |
Several of these can sit together. A product can keep Merge for the Common Model sync its customers expect, and send the one provider-specific read or write through Swytchcode, where policy, retries, and the audit log apply. More head-to-heads are in the comparison index.
How we checked these facts
Every price, catalog size, and feature claim in this article was checked against the vendor's own pricing page, documentation, or GitHub repository in November 2026. Where a vendor publishes two different numbers, both are named. Claims that come from a competitor's comparison are labeled with that source. Swytchcode facts come from docs.swytchcode.com and swytchcode.com/pricing.
FAQ
What is the closest alternative to Merge.dev?
Apideck and Unified.to are the closest if you want another unified API. Kombo and Finch are closer if your data is HR, ATS, or payroll. Nango is the closest if you would rather own the integration code. Match the job first, then the field coverage.
Is Swytchcode a Merge.dev replacement?
For governed execution of the provider call, yes. Merge's value is one normalized model across many customers' systems. Swytchcode's value is a governed call: the right method, managed auth, policy, retries with idempotency, and an audit trail. Teams that need both keep Merge for the normalized sync and use Swytchcode for the provider-specific calls the Common Model does not cover.
Why are people looking for Merge.dev alternatives in 2026?
The most common reason is a field the Common Model does not carry. Merge offers Field Mappings, Remote Data, and Authenticated Passthrough, but Remote Data and Passthrough are on Professional and Enterprise, and Remote Data refreshes only when a Common Model field changes. Others want pricing that is not per Linked Account, or their own code in place of a fixed schema.
What happens when a provider field isn't in Merge's Common Model?
You have four routes. Add it with a Field Mapping. Read the provider's raw payload with Remote Data. Call the provider through Merge with Authenticated Passthrough. Or call the provider outside Merge. The last two are raw provider calls, so check who enforces policy, retries the write safely, and logs it before you let an agent make them.
How much does a Merge.dev alternative cost?
Apideck Launch is $599 a month for 25 consumers. Unified.to Grow starts at $750 a month for 750,000 API calls. Finch Starter is $65 per connection a month. Nango is free for 10 connections, then $50 a month pay-as-you-go. Pipedream is free to build, then $99 a month on annual Startup. Composio Pro is $29 a month, and Arcade Team is $25 a month plus usage. Kombo and Paragon quote. Swytchcode is free forever for 100 live commands a month, $29 a month for 10,000, and $149 a month for a million. Merge itself is free for 3 Linked Accounts, then $650 a month for up to 10. Confirm the number on each pricing page before you buy. Figures move.
Swytchcode resources
- Swytchcode website: https://www.swytchcode.com
- Swytchcode docs: https://docs.swytchcode.com
- MCP quickstart: https://docs.swytchcode.com/quickstarts/getting-started/mcp/
- Idempotency guide: https://docs.swytchcode.com/guides/idempotency/
- Pricing: https://www.swytchcode.com/pricing
- Comparison index: https://www.swytchcode.com/compare
More content
How to Stop an AI Agent From Booking the Same Meeting Twice
A scheduling agent that checks availability and then books the slot has a gap between those two steps — and a retry can fall straight into it. Here's how Swytchcode closes that gap.
How to Stop an AI Agent From Creating Duplicate Records
A retried "create" call doesn't update an existing contact, ticket, or row — it makes a second one. Here's why that happens across CRMs, databases, and ticketing tools, and how Swytchcode prevents it.
How to Stop an AI Agent From Sending the Same Email Twice
Most email APIs have no idempotency key to catch a repeat send, which makes a duplicate email a harder problem than a duplicate charge. Here's how Swytchcode handles the gap.
