reprox-mcp
# reprox-mcp
`reprox-mcp` is a TypeScript MCP server that lets coding agents search persistent repository memory created by [cliper](https://www.npmjs.com/package/cliper-memory).
Reprox reads the persistent Local JSON memory created by Cliper and exposes it to MCP clients without inventing repository facts.
## Tools
| Tool | Use it when | Returns |
| --- | --- | --- |
| `search_memory` | You need context for a question or task. | Matching memories grouped by retrieval category. |
| `list_gaps` | You need known risks or incomplete areas. | All gap summaries, high severity first. |
| `get_gap` | You need full context for one gap. | The gap and every memory it references. |
| `get_related` | You need memories connected to a known memory. | Outgoing and incoming relationship neighbors. |
| `get_architecture` | You need a structural repository overview. | All architecture and repository memories. |
| `get_timeline` | You need change or release history. | Commit, release, and timeline memories, newest first. |
## Quickstart
Build and link or install the package, then ensure the target repository has been initialized:
```sh
cliper init
reprox mcp --path /path/to/repository
```
If repository memory is missing, the server responds with: `Repository memory is not initialized; run cliper init first.`
## Codex configuration
Add this to `~/.codex/config.toml`:
```toml
[mcp_servers.reprox]
command = "reprox"
args = ["mcp", "--path", "/absolute/path/to/repository"]
```
Omit `--path` to use the server process's current working directory.
## Development
```sh
npm install
npm run build
npm run pack-clean
```
TDQS
Scored across 6 tools
Each tool has a clear, distinct purpose: search for context, list/get gaps, get related memories, get architecture, and get timeline. Overlap with the generic search is acceptable because the specific getters are clearly scoped to particular memory types.
All tool names use snake_case and an imperative verb prefix like search_, list_, or get_. The only minor deviation is get_related, which uses an adjective rather than a clear noun object, but the pattern remains highly predictable.
With six tools, the server is well-scoped for a repository-context memory retrieval service. Each tool covers a meaningful access pattern without unnecessary redundancy or bloat.
The tool surface covers the full read-only retrieval lifecycle: broad search, specific gap lookup, relationship traversal, architecture overview, and timeline/history access. No significant missing operation is apparent for the stated purpose.