apidrifter vs Schemathesis
Schemathesis and apidrifter both send requests to a running API and compare the responses with its OpenAPI document, and that is where the overlap ends. Schemathesis is a property-based tester built on Hypothesis: it generates many inputs per operation, including invalid ones, chains operations into stateful workflows, and reports crashes and conformance failures with a curl reproducer for each. apidrifter sends one realistic request per operation with values you or an agent supply, and reports each mismatch with two fixes: one if the spec is right, one if the implementation is. It also diffs spec versions and MCP servers, which Schemathesis does not do. If you want to find bugs by volume, Schemathesis is the more mature tool. If you want to know where the document and the server disagree and which side to change, that is what apidrifter is for.
Feature by feature
| Feature | apidrifter | Schemathesis |
|---|---|---|
| Spec vs implementation verification | Yes. One planned request per operation; status, media type, headers and body are checked against the document. | Yes. Generated requests per operation, with status_code_conformance, content_type_conformance, response_headers_conformance and response_schema_conformance checks. |
| Property-based fuzzing | No. Values come from the spec's examples, a filled plan, or an agent. | Yes. Built on Hypothesis; positive, negative and all generation modes; stateful sequences that follow OpenAPI links. |
| Breaking-change diff between spec versions | Yes. 357 polarity-aware change rules; exit 1 in CI. | No. |
| MCP server snapshot and diff | Yes. Tools, resources, prompts and instructions, with description changes tracked. | No. |
| Agent-native MCP interface | Yes. apidrifter serve exposes the loop as nine MCP tools over stdio. | Not documented. |
| Which side to fix | Yes. Every finding carries a fix for the spec and a fix for the implementation. | No. Each failure names the check that failed and ships a curl reproducer. |
| GitHub Action | Yes, for spec diffs: changelog in the job summary, a fail-on threshold, optional PR comment. run works in any CI step via npx. | Yes. schemathesis/action@v3. |
| Hosted or paid tiers | None. The CLI is free; a hosted service is planned. | None. The Schemathesis.io service has been discontinued, according to the migration notes. |
| Spec formats | OpenAPI 3.x and MCP snapshots. Swagger 2.0 must be converted to OpenAPI 3 first. | OpenAPI 2.0, 3.0, 3.1 and 3.2; GraphQL; gRPC unary methods. |
| Reports | Text, JSON, Markdown. | CLI output plus JUnit, HAR, JSON, HTML and VCR via --report. |
| Language and runtime | TypeScript. Node 22.12 or newer; run with npx. | Python 3.11 or newer. CLI, pytest and unittest integration. |
| License | Apache-2.0 | MIT |
When to use Schemathesis instead
Reach for Schemathesis when:
- You want volume. Thousands of generated inputs per operation find edge cases that one realistic request never will.
- You need negative testing: proof that the API rejects invalid input, missing headers and undeclared methods.
- You need stateful sequences (create, read, delete) with use-after-free and resource-availability checks.
- The API is GraphQL or gRPC, or your team lives in pytest.
Use them together
They answer different questions, so running both is cheap and not redundant. apidrifter first, because it is fast and its output tells you what to edit; Schemathesis second, for depth.
# 1. Does the document describe the server? One request per operation, two fixes per finding.
npx apidrifter run openapi.yaml --base-url http://localhost:3000
# 2. Does the server hold up under generated input, including invalid input?
uvx schemathesis run openapi.yaml --url http://localhost:3000 --mode all
# 3. Did this change break clients of the previous spec?
npx apidrifter breaking openapi.before.yaml openapi.yamlBoth tools send real requests. Point them at a dev or staging server, not production.
Frequently asked questions
- Is apidrifter a replacement for Schemathesis?
- No. Schemathesis is a fuzzer that finds crashes and conformance failures by generating many inputs. apidrifter sends one realistic request per operation and reports which side to fix, and it also diffs spec versions and MCP servers. Use Schemathesis for depth and apidrifter for a fast, actionable contract check.
- Which one finds more bugs?
- Schemathesis, by volume: it generates thousands of inputs, including invalid ones, and runs stateful sequences. apidrifter finds the mismatches a single well-formed request exposes, which is what breaks generated clients and SDKs, and explains each with a fix for the spec and a fix for the implementation.
- Can both run in the same CI job?
- Yes. Both are command-line tools with non-zero exit codes on failure. Run apidrifter run against the preview server for the contract check, Schemathesis for the fuzz pass, and apidrifter breaking against the previous spec version for compatibility.
- Does either tool send my spec to a third party?
- Both run locally and only talk to the server you point them at. Schemathesis documents that its hosted service has been discontinued; apidrifter has no hosted service yet.
Sources
- Schemathesis README
- Schemathesis checks reference
- Schemathesis CLI reference
- Schemathesis migration notes
- apidrifter CLI reference and rules reference
Last reviewed . If something here about Schemathesis is out of date, open an issue and we will fix it.