context-mode
Verifiedby mksglu
Keeps raw tool output out of coding agents' context by running code and commands in a sandbox, indexing results for search, and restoring session state after compaction, via MCP.
- Access
- MCPCLI
- Auth
- None
- Skills
- 4
Skills
Sandboxed Execution
Runs code in 12 languages or processes files in an isolated subprocess with ctx_execute, letting only stdout enter the agent's context.
Indexed Knowledge Base
Chunks markdown or fetched URLs into a SQLite FTS5 index with ctx_index and ctx_fetch_and_index, then returns snippets via ctx_search.
Session Continuity
Tracks file edits, git operations, tasks, errors, and decisions in SQLite, and rebuilds working state after compaction or resume.
Batch Execution
Runs several commands and search queries in one ctx_batch_execute call, with optional concurrency for I/O-bound batches.
Related Agents
Headroom
by Headroom Labs
Compresses tool outputs, logs, RAG chunks, files, and history locally before they reach the model, as a Python or…
claude-mem
by thedotmack
Captures Claude Code tool usage as searchable observations and replays summarized context into new sessions, giving…
Impeccable
by Paul Bakaus
Gives coding agents a frontend design skill with 24 /impeccable commands, plus a CLI and Chrome extension that flag…
Trail-code
by Trail
Builds a local, always-fresh graph of your codebase as linked markdown and serves it to coding agents over MCP and a…