After update to v1.41.0 getting error: Script not found "stream-json"

[stefanpartheym]

I updated to meridian v1.41.0 and I'm now getting the following error whenever i try to send a prompt:

sdk_termination reason=process_exit exit=1 model=opus[1m] resume=false deferred=false stderr="error: Script not found \"stream-json\""

[rynfar]

do you have any more context for this by chance you can share? Anything you can tell me that lead up to this? Is it happening on every request now?

Which harness are you using? Opencode? Any telemetry you can share would be helpuful! @stefanpartheym

[stefanpartheym]

Hey @rynfar , yes sorry that was a bit brief.

• It is happening on every request (even with v1.40.0).
• I'm using OpenCode.
• Logs from meridian:

[
    {
        "level": "error",
        "category": "error",
        "message": "b0e73953-5bf1-4c9b-94b9-9c1061d05813 sdk_termination reason=process_exit exit=1 model=opus[1m] resume=false deferred=false stderr=\"error: Script not found \\\"stream-json\\\"\"",
        "requestId": "b0e73953-5bf1-4c9b-94b9-9c1061d05813",
        "timestamp": 1777913765421
    },
    {
        "level": "info",
        "category": "session",
        "message": "b0e73953-5bf1-4c9b-94b9-9c1061d05813 adapter=opencode model=opus[1m] stream=true tools=11 lineage=new session=new agent=primary active=1/10 msgCount=1",
        "requestId": "b0e73953-5bf1-4c9b-94b9-9c1061d05813",
        "timestamp": 1777913765416
    },
    {
        "level": "error",
        "category": "error",
        "message": "10128d26-4ecc-4a76-beb2-cd5fb5482080 sdk_termination reason=process_exit exit=1 model=opus[1m] resume=false deferred=false stderr=\"error: Script not found \\\"stream-json\\\"\"",
        "requestId": "10128d26-4ecc-4a76-beb2-cd5fb5482080",
        "timestamp": 1777913757402
    },
    {
        "level": "info",
        "category": "session",
        "message": "10128d26-4ecc-4a76-beb2-cd5fb5482080 adapter=opencode model=opus[1m] stream=true tools=11 lineage=new session=new agent=primary active=1/10 msgCount=1",
        "requestId": "10128d26-4ecc-4a76-beb2-cd5fb5482080",
        "timestamp": 1777913757397
    },
    {
        "level": "error",
        "category": "error",
        "message": "dc7a19f4-babc-46c2-9d73-c346d6a1830e sdk_termination reason=process_exit exit=1 model=opus[1m] resume=false deferred=false stderr=\"error: Script not found \\\"stream-json\\\"\"",
        "requestId": "dc7a19f4-babc-46c2-9d73-c346d6a1830e",
        "timestamp": 1777913753388
    },
    {
        "level": "info",
        "category": "session",
        "message": "dc7a19f4-babc-46c2-9d73-c346d6a1830e adapter=opencode model=opus[1m] stream=true tools=11 lineage=new session=new agent=primary active=1/10 msgCount=1",
        "requestId": "dc7a19f4-babc-46c2-9d73-c346d6a1830e",
        "timestamp": 1777913753384
    },
    {
        "level": "error",
        "category": "error",
        "message": "e4a5b367-66e2-44a1-8ac6-db019c99129d sdk_termination reason=process_exit exit=1 model=opus[1m] resume=false deferred=false stderr=\"error: Script not found \\\"stream-json\\\"\"",
        "requestId": "e4a5b367-66e2-44a1-8ac6-db019c99129d",
        "timestamp": 1777913751365
    },
    {
        "level": "info",
        "category": "session",
        "message": "e4a5b367-66e2-44a1-8ac6-db019c99129d adapter=opencode model=opus[1m] stream=true tools=11 lineage=new session=new agent=primary active=1/10 msgCount=1",
        "requestId": "e4a5b367-66e2-44a1-8ac6-db019c99129d",
        "timestamp": 1777913751347
    },
    {
        "level": "error",
        "category": "error",
        "message": "42dddf35-5341-40b2-8a30-b5f241f1faa3 sdk_termination reason=process_exit exit=1 model=haiku resume=false deferred=false stderr=\"error: Script not found \\\"stream-json\\\"\"",
        "requestId": "42dddf35-5341-40b2-8a30-b5f241f1faa3",
        "timestamp": 1777913751333
    },
    {
        "level": "info",
        "category": "session",
        "message": "42dddf35-5341-40b2-8a30-b5f241f1faa3 adapter=opencode model=haiku stream=true tools=0 lineage=new session=new agent=primary active=1/10 msgCount=1",
        "requestId": "42dddf35-5341-40b2-8a30-b5f241f1faa3",
        "timestamp": 1777913751321
    }
]

Error in OpenCode:

[rynfar]

thank you for the extra details, let me see if i can figure out whats going on here. @stefanpartheym

[rynfar]

Hey @stefanpartheym — thanks for the patience. I dug into this and I have a strong hypothesis, plus a small change I just shipped that should let us pin it down in one round trip.

What's most likely happening

The stderr error: Script not found "stream-json" is Bun's CLI error format — specifically what Bun emits when invoked as bun <args> (without run) and the first non-flag token isn't a runnable script. The SDK args start with ["--output-format", "stream-json", "--verbose", "--input-format", "stream-json"], so if Bun is the spawned process, it sees --output-format as an unknown flag and treats stream-json as a script name.

Translation: the executable Meridian's resolver picked is invoking Bun rather than the real Claude binary. Most likely culprit is a claude shim on your PATH that points at Bun (some bun-based tools install one), but it could also be MERIDIAN_CLAUDE_PATH set incorrectly or a partial install leaving a stub at the bundled location.

The frustrating part of the original issue: there was no way to tell which of those it was without playing detective on your filesystem. So I shipped a fix for that.

Self-diagnosis (after upgrade)

Once the next release lands (or if you want to try main now), the proxy logs the resolved Claude path + which step picked it on startup, and /health exposes the same:

curl http://127.0.0.1:3456/health | jq '.claudeExecutable'

You'll get something like:

{
  "path": "/Users/.../node_modules/@anthropic-ai/claude-code/bin/claude.exe",
  "source": "bundled"
}

Where source is one of:

Source	What it means
env	Picked up via MERIDIAN_CLAUDE_PATH
bundled	Found at node_modules/@anthropic-ai/claude-code/bin/claude.exe (the expected path)
platform-package	Fallback to @anthropic-ai/claude-code-<platform>-<arch>/claude
path-lookup	Picked from your PATH via which/where claude
legacy-cli-js	Old SDK fallback (Bun-only, very rare)

Can you upgrade and paste the /health output here? My bet is source: "path-lookup" and the path will tell us exactly which shim is in the way.

Immediate workaround (while waiting for the release)

Set MERIDIAN_CLAUDE_PATH to a known-good binary so the resolver short-circuits to it:

# Find Meridian's bundled binary (this is the canonical one)
MERIDIAN_CLAUDE_PATH="$(npm root -g)/@rynfar/meridian/node_modules/@anthropic-ai/claude-code/bin/claude.exe"

# Verify it actually runs
"$MERIDIAN_CLAUDE_PATH" --version
# Should print something like: 2.1.119 (Claude Code)

# Then export and restart meridian
export MERIDIAN_CLAUDE_PATH
meridian

If --version prints the version cleanly, that path is good and meridian will use it. If it errors instead, the bundled binary on your install is broken (separate class of issue, also pin-down-able via the new /health field).

Will leave this open until you've had a chance to grab the diagnostic — once we see the source tag this should be a fast fix.

[stefanpartheym]

Thanks for the update!

I think you missed some places:

• https://github.com/rynfar/meridian/blob/main/src/proxy/models.ts#L234
• https://github.com/rynfar/meridian/blob/main/src/proxy/profileCli.ts#L47

In my case, I don't have claude installed globally (I try to avoid it). Which then leads meridian to not be able to determine auth status. As soon as i add a small calude shim in my path meridian profile list gives me proper status again:

#!/usr/bin/env bash
"$HOME/.config/opencode/node_modules/.bin/claude" "$@"

The meridian server running - for me - is a different story. I'm running it as a systemd service. Even with a claude shim, it still fails.
Logs:

Mai 06 15:00:00 fux systemd[4818]: Started Meridian - Anthropic API proxy.
Mai 06 15:00:00 fux bun[370369]: ⚠ Could not verify Claude auth status. If requests fail, run: claude login
Mai 06 15:00:00 fux bun[370369]: [PROXY] Plugins loaded: 1 active, 0 disabled, 0 errors
Mai 06 15:00:00 fux bun[370369]: Meridian running at http://127.0.0.1:3456
Mai 06 15:00:00 fux bun[370369]: Telemetry dashboard: http://127.0.0.1:3456/telemetry
Mai 06 15:00:00 fux bun[370369]: Model pins: opus=claude-opus-4-7 sonnet=claude-sonnet-4-6 haiku=claude-haiku-4-5
Mai 06 15:00:00 fux bun[370369]: Claude executable: /tmp/bunx-1000-@rynfar/meridian@latest/node_modules/@anthropic-ai/claude-code-linux-x64/claude (resolved via platform-package)
Mai 06 15:00:00 fux bun[370369]: Point any Anthropic-compatible tool at this endpoint:
Mai 06 15:00:00 fux bun[370369]:   ANTHROPIC_API_KEY=x ANTHROPIC_BASE_URL=http://127.0.0.1:3456

[rynfar]

No problem! Thanks for the additional context this will help us get to the bottom of it. I will take another pass at this.

[rynfar]

@stefanpartheym — your pointers were exactly right. There were five places (not two) where Meridian was still spawning `claude` via shell PATH instead of routing through the resolver:

• `bin/cli.ts:157` (the boot pre-flight that produces the "Could not verify Claude auth status" warning you saw)
• `src/proxy/models.ts:234` (per-request auth check — your pointer)
• `src/proxy/profileCli.ts:47` (`meridian profile list` — your other pointer)
• `src/proxy/profileCli.ts:136` (`meridian profile add` browser flow)
• `src/proxy/profileCli.ts:306` (`meridian profile login` browser flow)

All five now route through a new `resolveClaudeExecutableSync` helper (sync subset of the resolver, covering env / bundled / platform-package — the steps that don't need async exec) and use `execFile`/`execFileSync`/`spawnSync` with the resolved path directly. No more shell PATH dependency.

PR is up at #492. End-to-end verified by booting the patched build with `PATH=/usr/local/bin:/usr/bin:/bin` (no `claude` shim anywhere on PATH) — no warning at startup, `/health` reports `loggedIn: true` cleanly.

Once it merges + releases (likely 1.42.1), the `claude` shim you added as a workaround should no longer be necessary, and your systemd setup should work without modification. Will ping you here when the release is out so you can confirm on your machine.

Thanks again for the precise diagnosis — saved a lot of guessing.

[stefanpartheym]

Hey @rynfar ! Thank you. I can confirm it works now :)
Claude executable is now resolved properly: "claudeExecutable":{"path":"/tmp/bunx-1000-@rynfar/meridian@^1.42.1/node_modules/@anthropic-ai/claude-code-linux-x64/claude","source":"platform-package"}