Runs chimp's MCP client against the official MCP conformance test suite.

The conformance harness is the inverse of a server: it starts a test MCP server, spawns the binary configured via --command (this fat-jar wrapped in bin/chimp-conformance-client), and passes the test-server URL as the last argument. The client process reads the MCP_CONFORMANCE_SCENARIO env var to decide what protocol exchange to drive, talks to the harness's server, and exits 0 on success.

This subproject is the binary the harness invokes. Main.scala dispatches on the scenario name and selects the transport per scenario:

The harness also sets MCP_CONFORMANCE_PROTOCOL_VERSION; Main.scala uses it to select the protocol version. The sbt task sets CHIMP_CLIENT_CONFORMANCE_JAR to the path of the assembled fat jar; the wrapper script uses it, and falls back to a glob over target/scala-* when the variable is not set.

The sbt task pins the harness to the version set in conformanceHarnessV in build.sbt.

Run the full requirement set:

sbt 'clientConformance/conformance client --requirements 2025-11-25'

Run one scenario during development:

sbt 'clientConformance/conformance client --scenario initialize --spec-version 2025-11-25'
sbt 'clientConformance/conformance client --scenario tools_call --spec-version 2025-11-25'

The sbt task downloads @modelcontextprotocol/conformance with npm; npx must be available on the PATH.

Add a case in Main.scala matching on the scenario name (the value of MCP_CONFORMANCE_SCENARIO), drive the protocol with McpClient, return exit code 0 on success. Once it passes, remove the entry from the baseline file (see below).

conformance-baseline.yml lists scenarios that are known to fail today. The harness uses it like this:

So the file shrinks as the SDK matures.