MCP server exposing a long-lived interactive REPL runtime (R or Python backend) over stdio.
Wellknown found it in public sources; nobody has proven control of it yet. Claiming takes one click if the repository is under your GitHub account, or a small file on your domain otherwise. Verified owners get the badge, 15-minute checks, status alerts, edits that outrank crawled data, and a ranking boost.
Agents can do it too: POST https://wellknown.network/api/v1/claims with {"agent":"posit-mcp-repl","method":"well_known_file"} — machine-readable steps at claim.json, guide at /docs/claim.
Everything here was measured by our prober or read from a registry. Nothing is self-reported.
Attributed to the source that supplied each field. Treated as claims, not facts.
# mcp-repl `mcp-repl` is an MCP server that gives an agent a persistent Python or R session, kept alive across tool calls. The agent can load data once, inspect objects, try ideas, read help, make plots, and keep iterating — the way a person would in a REPL. A shell tool running `Rscript -e` or `python -c` keeps forcing the agent to rebuild context. `mcp-repl` keeps the session open instead: variables, loaded packages, plots, and other state stay available until you or the model reset. ## Features - **Sandboxed by default.** The backend runs in a sandbox enforced by OS primitives at the process level — not command-specific runtime rules. Network is disabled; writes are constrained to workspace roots and the temp paths the active session needs. On Unix, a memory guardrail kills the worker if it exceeds threshold. - **Curated output.** Smart echo (omitted when safe, elided for large multi-expression blocks) and in-band help pages. Plots are returned as inline images through MCP for vision-capable models, so the agent sees the plot directly; non-vision models still get the saved file path. When replies get too large, the tool response stays short and the full output is saved as a structured bundle (transcript + plot files) the model can explore on demand. - **No polling.** R and Python run embedded in the worker, not behind a stdio pipe driven by prompt-string heuristics. The server knows precisely when the interpreter is idle and has settled, so each `repl` call returns the moment the work is done — no fixed waits, no guessing whether more output is on the way. - **Explicit session control.** Interrupts and resets are first-class. ## Quickstart ### 1. Install Install from PyPI. The package is named `posit-mcp-repl` and exposes the `mcp-repl` executable, plus a `posit-mcp-repl` alias for `uvx`: ```sh pipx install posit-mcp-repl # or uv tool install posit-mcp-repl # one-off uvx posit-mcp-repl --help ``` Or install via `cargo` (needs…
Mapped onto the structured taxonomy from declared text and observed tool names. Confidence shown for derived entries.
Every source is kept verbatim. Field changes are logged as events.