harness-mcp-server
v0.1.3
Published
Recursive Language Model and Continual Harness in one MCP server: long-context Q&A, a Python kernel, and durable memory.
Downloads
531
Maintainers
Readme
harness-mcp-server
The Recursive Language Model and the Continual Harness in one MCP server. It serves every tool from both, so a host gets long-context Q&A, a persistent Python kernel, and durable memory over a single connection.
Part of the harness monorepo. Use this when you want one process; the two capabilities also ship as separate servers, rlm-mcp-server and continual-harness-mcp-server.
Quick start
npm install
npm run build
npm run dev # stdio server for a local hostOr run it without cloning:
npx -y harness-mcp-serverThe installed command is harness, which is also a CLI that routes to each side:
harness rlm query "who led it?" --file report.txt
harness continual add-memory "Prefer focused patches" "Small updates validate faster."No subcommand starts the combined MCP server, so hosts are unaffected.
Set RLM_MODEL to a provider/model string such as anthropic/claude-sonnet-5 to
enable answering, and RLM_SANDBOX=local to enable the Python kernel. Harness state
follows the same paths as continual-harness-mcp-server.
Tools
The union of both servers: rlm_store_context, rlm_query, python_exec,
rlm_deep_query from the RLM, and harness_get_state, harness_create_memory,
harness_create_prompt, harness_create_skill, harness_create_subagent,
harness_update, harness_delete, harness_refine from the harness. See each
package's README for detail.
