The short version
An API is a contract between two programs: the developer decides, at design time, which requests to make and what to do with the responses. MCP is a protocol between an AI application and a server: the model decides, at run time, which tools to call based on the question in front of it. They solve different problems and most serious deployments use both.
Side by side
| Dimension | Company Data API | DataGardener MCP |
|---|---|---|
| Who decides the calls | Your developers | The agent, from the question |
| Typical use | Onboarding checks, enrichment jobs, scheduled syncs | Briefs, due diligence, market questions, monitoring with reasoning |
| Interface | REST endpoints and documentation | Task-level tools discovered by the host |
| Output | Fields | Fields plus provenance, usage and a narrative |
| Latency per answer | One request | Several tool calls |
| Change management | Code change per new use case | New question, no code change |
| Entitlement | API key scoped to a customer | OAuth connection scoped to a customer |
When the API is the right choice
- The flow is fixed and must be deterministic (compliance checks, batch enrichment).
- Latency and cost per call matter more than flexibility.
- No model is involved, or the model only formats a known result.
When MCP is the right choice
- The question is open and the path to the answer depends on what is found.
- Users ask in natural language and expect the evidence to be shown.
- You want new use cases without new integration code.
Does MCP replace APIs?
No. An MCP server usually calls APIs internally. MCP standardises discovery and invocation for AI applications; APIs remain the backbone for programmatic integration. The DataGardener Company Data API and DataGardener MCP serve the same intelligence through two interfaces.
Trade-offs stated fairly
Agents make more calls than a scripted integration and need governance: tool permissions, entitlement and usage visibility. That is why DataGardener MCP tools are read-only, entitled per connection and report usage. For fixed, high-volume flows, the API remains the efficient path.