Start with scope, not the biggest red field
REA doctor reports more than whether an MCP registration exists. Read the selected scope, the overall readiness fields, the individual checks and their remediation together. A missing analysis engine is not automatically evidence that a client failed to connect.
Run doctor for the client and workflow you actually intend to use:
npx -y rea-agents@6.7.0 doctor --client claude_code --skill --json
If you installed into a separate profile, prefix this command with the same CLAUDE_CONFIG_DIR used during setup. A default-profile diagnosis cannot establish the state of a different profile. Do not paste the complete output publicly without checking paths, environment details and registration values for sensitive information.
A practical decision table
| What you observe | What it establishes | Next check |
|---|---|---|
| Unsupported Node remediation | REA's runtime prerequisite failed | Update the runtime used by the command, then rerun |
| Missing registration in the selected client | Doctor did not find the expected registration in that scope | Compare the profile and config path with setup |
| Skill state missing or version mismatch | Workflow files are absent or out of sync | Preview setup for that same client |
| Missing analysis engine | That provider is not ready for its tasks | Configure only the provider your task needs |
| Selected readiness passes | The selected checks passed | Verify connection in the actual client |
| Client connects, model rejects a schema | Transport and model compatibility are separate | Read the schema error, active version and gateway details |
The last step is not optional if your goal is an actual analysis. An entry in a JSON file is a launch instruction. A connection result is a running integration. A useful tool result is another milestone.
Why can environment_healthy be false?
The full environment view can include providers you have not configured. In the isolated 6.7.0 test, doctor returned healthy: true, environment_healthy: false, with the skill state aligned. Provider inspection reported Ghidra, IDA, jadx and Kuna configuration as missing. Those checks describe engine availability; they do not by themselves prove that the Claude Code registration is malformed.
Read the check's classification, code, detail and remediation. For example, missing_analysis_engine with not_configured is different from an invalid environment map or a malformed client file. Do not turn every optional-engine warning into a requirement to install a large reverse-engineering suite.
For a specific task, inspect the relevant provider. The official guide documents provider-scoped checks such as:
npx -y rea-agents@6.7.0 doctor --provider ghidra --json
Use this only if Ghidra is the engine you mean to use. Read its version and path requirements in the current installation documentation. A path merely existing on disk is not the same as a provider passing its readiness checks.
A reproduced runtime failure
Our first dry-run used Node 24.5.0 and returned status: needs_human with no applied actions. Its remediation requested a supported Node version. Under Node 24.11.1, the same isolated setup produced a plan, then applied the client registration and skill installation successfully.
This example is useful because the diagnostic output also listed missing engines. Installing engines first would have addressed a different issue. Diagnose the immediate blocker, rerun, and then decide which remaining checks are relevant to the task.
Tools still do not show up
Check the exact client process and profile. Restart after a registration update. Compare the configured package version with the version you invoked manually. A registration pinned to an old package can continue to start it even after a newer CLI was downloaded elsewhere.
Then inspect the client's MCP status and logs. Do not infer a successful connection from doctor alone. If tools are present but calls fail, retain the tool name and a sanitized error category. Do not publish tokens, binary contents or confidential file paths in an issue report.
What to include in a useful bug report
Record the REA version requested by the registration, the version actually started if the client exposes it, Node version, client name/version, operating system, selected scope and the first failed check. State whether setup, client connection or an individual tool call failed. These distinctions let a maintainer reproduce the same layer.
For provider schema rejection, also record the model gateway and whether compact mode changed the outcome; see REA schema errors. For an initial installation, return to REA setup. For JSON-only mistakes, the local checker can identify duplicate members and incorrect field types without launching anything.
Source: REA installation and doctor documentation. Our checks do not cover licensed engines, paid provider requests or every client platform.