docs: add a DeepSeek Harness overlay example - #34
Merged
Conversation
dsh ships no memory layer, and its design notes reject per-provider plugins
outright — "it repeats auth, configuration, lifecycle, and tool wrappers that
MCP already standardizes and expands ownership for every added provider". The
supported path is a Cordis overlay aiming their generic MCP client at whatever
server you want.
So there is nothing to build. mneme already speaks stdio MCP, which is the only
thing that overlay requires; the integration is a config file. Adding it here
rather than upstream because dsh does not accept external pull requests.
Structure mirrors their examples/mcp-memory overlays, with one deviation worth
naming: `command: node` plus an absolute script path instead of a bare
executable, because mneme is not published to npm with a `bin` and there is
nothing on PATH to invoke.
The comments carry the two failure modes a new user actually hits, both of
which are silent rather than loud:
- no TOKENMEM_DB_PATH -> DB resolves against the process working directory,
so the store built in one session can be invisible from the next
- no EMBEDDING_API_* -> starts and answers fine, as FTS5 keyword search only.
Hybrid recall is the reason to run mneme; its absence is not an error, just
a quieter and worse result set
README lists it under "Config provided, not verified by us" rather than in the
"Tested with" table. The CI here exercises mneme's own MCP server, not dsh's
loader, and claiming otherwise in the same document that warns about silent
degradation would be its own small version of the same sin.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why there is no plugin here
dsh ships no memory layer of its own, and its design notes reject per-provider plugins outright:
The supported path is a Cordis overlay pointing their generic
@deepseek-ai/dsh-mcp-clientat whatever server you want. mneme already speaks stdio MCP — which is the only thing that overlay requires — so the integration is a config file and nothing else.Added here rather than upstream because dsh does not accept external pull requests ("We are sorry that we cannot accept external pull requests at the moment").
What's in it
examples/dsh.cordis.yml, structured to mirror their ownexamples/mcp-memoryoverlays. One deviation:command: nodeplus an absolute script path instead of a bare executable, because mneme is not on npm with abinand there is nothing on PATH to invoke.The comments carry the two failure modes a new user actually hits, both silent:
TOKENMEM_DB_PATHEMBEDDING_API_*Honesty about coverage
README lists this under "Config provided, not verified by us", not in the "Tested with" table. CI here exercises mneme's own MCP server, not dsh's loader — and claiming otherwise, in the same document that warns about silent degradation, would be its own small version of that sin.
Docs + one example file. No code paths touched.
🤖 Generated with Claude Code