Maintainer-run acceptance check performed on 2026-08-08. This checks the OpenCode/MCP integration boundary, not production-model quality or independent user adoption.
- macOS 15.6.1 (arm64)
- OpenCode 1.18.15, invoked as
npx -y [email protected] - GoodMemory 0.7.2, invoked as
npx -y [email protected] - Node.js 22.14.0
- Bun 1.3.14
- Isolated HOME, XDG config/data/cache, SQLite database, and two synthetic projects
- Deterministic local OpenAI-compatible test model
--autowas used only inside the disposable test environment so the model could invoke MCP tools non-interactively
The published artifacts were verified immediately before the run:
- npm version:
0.7.2 - npm tarball:
https://registry.npmjs.org/goodmemory/-/goodmemory-0.7.2.tgz - GitHub release:
https://github.com/hjqcan/GoodMemory/releases/tag/v0.7.2
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"goodmemory": {
"type": "local",
"command": [
"npx",
"-y",
"[email protected]",
"mcp",
"serve",
"--standalone",
"--user-id",
"opencode-e2e",
"--allow-write",
"--storage-url",
"/tmp/goodmemory-opencode-e2e/goodmemory.sqlite"
],
"enabled": true
}
}
}The test model was configured as a custom OpenAI-compatible provider. The full
deterministic server and opencode.json used for the run are included in this
gist.
opencode mcp list reported:
✓ goodmemory connected
npx -y [email protected] mcp serve --standalone --user-id opencode-e2e --allow-write --storage-url /tmp/goodmemory-opencode-e2e/goodmemory.sqlite
The model request exposed exactly these nine GoodMemory tools:
goodmemory_get_context
goodmemory_get_records
goodmemory_inspect_memory
goodmemory_read_artifacts
goodmemory_remember
goodmemory_search_index
goodmemory_stats
goodmemory_timeline
goodmemory_trace_recall
OpenCode prefixes them as goodmemory_goodmemory_* because the configured MCP
server name is goodmemory.
The model invoked goodmemory_goodmemory_remember with:
{
"content": "The Orion service deployment strategy is blue-green deployment.",
"cwd": "/tmp/goodmemory-opencode-e2e/project-a",
"role": "user"
}The tool result contained:
{
"accepted": 1,
"memoryIds": ["baa41cc9-ed8d-42ca-bb81-9955e88a10a3"],
"scope": {
"userId": "opencode-e2e",
"workspaceId": "project-a"
},
"storage": { "provider": "sqlite" }
}The SQLite document store contained one facts record whose content and scope
matched the tool result.
A separate opencode run created a different session ID and invoked
goodmemory_goodmemory_get_context for the Orion deployment question. The tool
returned:
Evidence: - The Orion service deployment strategy is blue-green deployment.
Facts: - The Orion service deployment strategy is blue-green deployment.
The returned scope was opencode-e2e / project-a.
The same GoodMemory SQLite database and user ID were reused from
/tmp/goodmemory-opencode-e2e/project-b. goodmemory_get_context returned only
the empty heading Developer memory notes: with scope
opencode-e2e / project-b; it did not return the Orion fact.
Back in project A, goodmemory_trace_recall returned one fact candidate with:
{
"returned": true,
"lexicalScore": 0.6666666666666666,
"reason": "scope_match",
"sourceMethod": "explicit",
"scope": {
"userId": "opencode-e2e",
"workspaceId": "project-a"
}
}The trace also linked the evidence record and reported deterministic
rules-only routing because semantic search was unavailable in this isolated
run.
This proves current package compatibility, MCP tool discovery, an explicit write, cross-session recall, project isolation, and recall tracing in OpenCode 1.18.15 on this environment. It does not prove automatic capture, native OpenCode hypothesis-loop detection, productivity improvement, Windows/Linux support, production-model behavior, or independent-user adoption.