Claude and MCP

Jers ships as an MCP server and a Claude Code plugin, so Claude can make typed decisions with memory as tool calls.

Claude Code

export JERS_API_KEY=jj_live_...
export JERS_BASE_URL=https://api.getjers.com   # the default
export JERS_MODEL=jers-english                 # optional: the model decide uses when a call names none

The plugin is not published yet. When it is, adding it takes two commands in Claude Code (/plugin marketplace add, then /plugin install jers@jers); until then, call the API directly.

The plugin brings the tools and a skill on writing questions and memory lines that work.

Tools

ToolDoes
decideanswers typed questions about a state; model, subject, compare (with a subject), robust, values, derive
rememberstores facts about a subject, with write-time warnings
memorywhat Jers remembers about a subject, word for word
forgetremoves a concept and checks recall
deleteremoves a subject's whole memory
feedbackthe right answer for an earlier decision
qualityaccuracy and calibration from that feedback

Each successful result has a short text for the model and the full API response as structuredContent (read by clients on MCP 2025-06-18); an error result has the text only. Windows, the cache, top_k, min_share, placebo, rules, golden sets, batches and calibration are not tools: use the API or an SDK for those. decide does fire the rules a subject already has.

Other MCP clients

The server is one file of standard-library Python (plugins/jers/server/jers_mcp.py) speaking MCP over stdio, versions 2025-06-18, 2025-03-26 (including batches of messages) and 2024-11-05:

{"mcpServers": {"jers": {"command": "python3", "args": ["/path/to/plugins/jers/server/jers_mcp.py"],
                         "env": {"JERS_API_KEY": "jj_live_...", "JERS_BASE_URL": "https://api.getjers.com", "JERS_MODEL": "jers-english"}}}}

Without a key it still starts, and every tool call says how to get one.