Local-first context-engineering MCP server and CLI that gives Claude token-efficient file reads, shell-output compression, code search, graph queries, persistent session memory, context packaging, verification tools, and dashboard-style token accounting through a single Rust binary.
LeanCTX can read local files, run shell commands invoked through its tools, cache outputs, install shell/editor hooks, and persist session state., Review generated MCP and shell-hook changes before enabling them across all agents or shells., Keep path jail enforcement enabled for normal use, and allow extra roots only when a project genuinely needs them., Disable or restrict command-execution and unsafe I/O tools for regulated repositories, untrusted workspaces, or shared team environments., Treat compressed shell output and cached reads as summaries; switch to full reads or raw command output when exact source text matters.
Privacy notes
Local code, file paths, command output, search terms, session notes, context packages, knowledge graph data, token metrics, and dashboard statistics can be sent to the MCP client and model., LeanCTX stores local stats and session state under its own configuration/data directories; protect these files if they contain project decisions, findings, or sensitive paths., Secret-like paths are blocked or role-gated by default according to upstream security docs, but users should still avoid prompting agents to read credentials, private keys, tokens, or ignored files., The upstream security policy describes optional update checks and opt-in anonymous stats sharing; disable network checks when working in confidential or offline environments.
Author
Yves Gude
Submitted by
oktofeesh1
Claim status
unclaimed
Last verified
2026-06-06
Decision playbook
Review trust signals before you adopt
Signals are present but mixed. Use the checklist below to confirm the source and operational safety for your environment.
Compare context
Selected
0
Current score
63
Baseline
—
Delta
No baseline selected
No major trust-signal divergence detected in the current selection.
Source and provenance checks
Needs review
Confirm ownership and provenance before trusting install instructions.
Source link availableRequired
Open the canonical repository and verify ownership.
Done
Source provenance statusRequired
Marked as source-backed.
Done
Metadata reviewed
No reviewed flag detected in metadata.
Pending
Safety and privacy checks
Complete
Validate risk disclosures before installation or API wiring.
Safety notes presentRequired
Review the listed safety guidance before running commands.
Done
Privacy notes presentRequired
Review data handling notes before connecting accounts or secrets.
Done
Trust level risk gateRequired
Trust level does not block evaluation.
Done
Package and install checks
Needs review
Check package metadata and artifact integrity signals.
Install payload available
Install or copy payload is available for review.
Done
Package verification flag
No package verification flag provided.
Pending
Checksum metadata
No checksum provided for downloaded artifact.
Pending
Compare-driven decision checks
Needs review
Use compare context to validate trade-offs before adoption.
Compare tray has multiple entries
Add at least one more entry to compare trust differences.
5 safety and 4 privacy notes across 4 risk areas. Review closely: credentials & tokens, network access.
4 areas
SafetyCredentials & tokensLeanCTX can read local files, run shell commands invoked through its tools, cache outputs, install shell/editor hooks, and persist session state.
SafetyExecution & processesReview generated MCP and shell-hook changes before enabling them across all agents or shells.
SafetyLocal filesKeep path jail enforcement enabled for normal use, and allow extra roots only when a project genuinely needs them.
SafetyExecution & processesDisable or restrict command-execution and unsafe I/O tools for regulated repositories, untrusted workspaces, or shared team environments.
SafetyExecution & processesTreat compressed shell output and cached reads as summaries; switch to full reads or raw command output when exact source text matters.
PrivacyCredentials & tokensLocal code, file paths, command output, search terms, session notes, context packages, knowledge graph data, token metrics, and dashboard statistics can be sent to the MCP client and model.
PrivacyCredentials & tokensLeanCTX stores local stats and session state under its own configuration/data directories; protect these files if they contain project decisions, findings, or sensitive paths.
PrivacyCredentials & tokensSecret-like paths are blocked or role-gated by default according to upstream security docs, but users should still avoid prompting agents to read credentials, private keys, tokens, or ignored files.
PrivacyNetwork accessThe upstream security policy describes optional update checks and opt-in anonymous stats sharing; disable network checks when working in confidential or offline environments.
Safety notes
LeanCTX can read local files, run shell commands invoked through its tools, cache outputs, install shell/editor hooks, and persist session state.
Review generated MCP and shell-hook changes before enabling them across all agents or shells.
Keep path jail enforcement enabled for normal use, and allow extra roots only when a project genuinely needs them.
Disable or restrict command-execution and unsafe I/O tools for regulated repositories, untrusted workspaces, or shared team environments.
Treat compressed shell output and cached reads as summaries; switch to full reads or raw command output when exact source text matters.
Privacy notes
Local code, file paths, command output, search terms, session notes, context packages, knowledge graph data, token metrics, and dashboard statistics can be sent to the MCP client and model.
LeanCTX stores local stats and session state under its own configuration/data directories; protect these files if they contain project decisions, findings, or sensitive paths.
Secret-like paths are blocked or role-gated by default according to upstream security docs, but users should still avoid prompting agents to read credentials, private keys, tokens, or ignored files.
The upstream security policy describes optional update checks and opt-in anonymous stats sharing; disable network checks when working in confidential or offline environments.
Prerequisites
npm for the `lean-ctx-bin` package, or another upstream-supported install path such as release binaries, Homebrew, Cargo, or AUR.
A local repository workspace where LeanCTX is allowed to read files, run configured shell commands, and store session metadata.
Review of generated MCP client config, shell hooks, per-project `.lean-ctx.toml`, and global `~/.config/lean-ctx/config.toml`.
Explicit allow-listing for any paths outside the current project root that LeanCTX should be permitted to access.
LeanCTX MCP Server connects Claude and other coding agents to a local
context-engineering layer. The server exposes ctx_* tools for compact file
reads, shell-output compression, code search, code graph analysis, context
packages, persistent session memory, multi-agent handoffs, token accounting,
verification, and context-budget control.
The project is local-first and runs as the user. Its security documentation
describes project-root path jailing, explicit allow roots, secret-like path
checks, output caps, deterministic redaction, JSON-RPC size limits, optional
update checks, and opt-in anonymous stats sharing.
These sources were reviewed on 2026-06-06. Prefer the live repository,
README, getting-started docs, npm registry metadata, release page, stdio
transport implementation, tool registry, generated MCP tool reference, security
policy, and license for current install paths, tool counts, transport behavior,
security boundaries, optional network activity, and licensing.
Features
Local Rust binary distributed through release archives, npm lean-ctx-bin,
Homebrew, Cargo, AUR, and other upstream-supported install paths.
MCP server configured by lean-ctx onboard or lean-ctx setup.
Stdio transport implementation for MCP clients.
Context tools for file reads, directory maps, code search, deltas, shell
compression, cache operations, token accounting, and context retrieval.
Graph and architecture tools for symbols, call graphs, impact analysis,
route discovery, code smells, and project overviews.
Session memory, knowledge, task, handoff, multi-agent, and workflow tools.
Context package, verification, proof, feedback, metrics, and budgeting tools.
Security controls for project-boundary path jailing, allow roots, secret-like
paths, output caps, redaction, message limits, and role-based restrictions.
Review the generated MCP client config, shell hooks, and per-project settings
before enabling LeanCTX across all AI clients. Run lean-ctx doctor after setup
and restart your shell and AI tool once so MCP tools and hooks load cleanly.
Use Cases
Let Claude read large source files through compact map, signature, diff,
task, or cached full modes.
Compress noisy shell output from git, npm, cargo, Docker, kubectl, Terraform,
and similar tools before it reaches the model.
Search a codebase, inspect symbols, trace call graphs, and estimate impact
without repeatedly dumping full files into context.
Persist task facts, decisions, handoffs, and session memory across long
coding sessions.
Build context packs for pull requests, reviews, debugging, or multi-agent
handoffs.
Track token savings, context utilization, and tool-level cost attribution.
Safety and Privacy
LeanCTX runs locally, but it sits directly between an AI agent and the user's
repository. Treat it as a privileged developer tool. Its MCP tools can read
files, summarize directories, search code, compress command output, persist
session data, and execute shell commands that the user or agent invokes.
Keep project-boundary enforcement on unless you intentionally allow more roots.
Review role settings before enabling unsafe I/O, ignored-file search, command
execution, team-server access, or context-package export. For sensitive
repositories, disable optional update checks and any opt-in stats sharing, and
store session data and context packages under the same controls as source code.
Even compressed outputs can disclose source code, secrets, filenames, incident
details, architecture, customer names, or unreleased product plans if sent to an
MCP client and model. Do not ask the agent to read credentials, private keys,
tokens, .env files, ignored files, or unrelated directories unless that access
is explicitly approved and scoped.
Duplicate Check
Existing MCP content includes code-search, repository-context, and coding-agent
helpers, but no dedicated LeanCTX, lean-ctx, lean-ctx-bin, or
yvgude/lean-ctx MCP entry was found in content/mcp. LeanCTX MCP Server is
distinct because it documents the local Rust binary's context compression,
stdio MCP server, code graph, session memory, shell-output compression,
verification, and token-accounting workflows.
Show that LeanCTX MCP Server is listed on HeyClaude. Paste this Markdown into your README — it renders the badge and links back to this page.
[](https://heyclau.de/entry/mcp/leanctx-mcp-server)
How it compares
LeanCTX MCP Server side by side with 3 alternatives on trust, install, platform support, and disclosed safety notes — all from reviewed registry metadata.
Local-first context-engineering MCP server and CLI that gives Claude token-efficient file reads, shell-output compression, code search, graph queries, persistent session memory, context packaging, verification tools, and dashboard-style token accounting through a single Rust binary.
High-performance MCP server that indexes codebases into a persistent knowledge graph for structural search, call tracing, architecture summaries, dead-code detection, and cross-repo analysis.
Stdio MCP server that lets Claude ask the Google Gemini CLI for large-file, codebase, brainstorming, and sandboxed analysis while preserving MCP tool and prompt workflows inside Claude Code or compatible MCP clients.
✓LeanCTX can read local files, run shell commands invoked through its tools, cache outputs, install shell/editor hooks, and persist session state.
Review generated MCP and shell-hook changes before enabling them across all agents or shells.
Keep path jail enforcement enabled for normal use, and allow extra roots only when a project genuinely needs them.
Disable or restrict command-execution and unsafe I/O tools for regulated repositories, untrusted workspaces, or shared team environments.
Treat compressed shell output and cached reads as summaries; switch to full reads or raw command output when exact source text matters.
✓Codebase Memory MCP reads source files across indexed repositories and writes MCP entries, instruction files, skills, hooks, and agent configuration files during install.
The background watcher and auto-index features can keep graph data updated as source files change.
The optional graph UI exposes a local visualization server on a localhost port.
Shared graph artifacts and SQLite databases can persist source metadata after an MCP session ends.
Verify downloaded release binaries, checksums, signatures, and provenance before running them in sensitive environments.
✓Context+ can inspect source files, build AST and embedding indexes, run static analysis, create memory graph nodes, and propose code changes.
The `propose_commit` tool is designed to write code after validation and creates shadow restore points before saving.
Static analysis tools may invoke local compilers, linters, or language tools depending on the repository.
Runtime caches and memory graph data can persist source-derived embeddings and relationships after the MCP session ends.
External embedding providers can receive source-derived text, identifiers, prompts, and cluster-labeling context.
✓Gemini MCP Tool runs locally but invokes the Gemini CLI, so prompts, file references, and command output can leave the local machine through the configured Google Gemini account or API path.
The `ask-gemini` tool can pass `@` file or directory references to Gemini CLI; current source checks that references stay under the working directory, but users should still scope prompts carefully.
Sandbox mode is exposed as an option on `ask-gemini` and forwards the Gemini CLI sandbox flag; review generated scripts, network calls, package installs, and filesystem changes before relying on sandboxed output.
Change mode can generate structured edit suggestions for Claude to apply; inspect proposed OLD/NEW replacements before applying them to source files.
Treat the project as an unofficial third-party bridge, not an official Google MCP server.
Privacy notes
✓Local code, file paths, command output, search terms, session notes, context packages, knowledge graph data, token metrics, and dashboard statistics can be sent to the MCP client and model.
LeanCTX stores local stats and session state under its own configuration/data directories; protect these files if they contain project decisions, findings, or sensitive paths.
Secret-like paths are blocked or role-gated by default according to upstream security docs, but users should still avoid prompting agents to read credentials, private keys, tokens, or ignored files.
The upstream security policy describes optional update checks and opt-in anonymous stats sharing; disable network checks when working in confidential or offline environments.
✓Source code, filenames, paths, symbols, comments, imports, routes, call graphs, architecture summaries, ADRs, graph queries, prompts, and tool outputs may be visible to the MCP client and model provider.
Persistent graph databases and shared artifacts can retain proprietary architecture, internal package names, route names, service boundaries, and accidentally committed secrets.
Auto-indexing broad workspace roots can include unrelated private repositories or generated files.
Avoid committing shared graph artifacts unless the repository owners explicitly approve storing derived code intelligence in version control.
✓Source code, file paths, symbol names, comments, identifiers, embeddings, memory nodes, relations, prompts, API keys, static-analysis output, proposed diffs, and tool results may be visible to the MCP client and model provider.
Proprietary codebases can expose internal architecture, feature maps, naming conventions, secrets accidentally present in code, and product plans.
Keep API keys and provider base URLs out of committed MCP configs, logs, screenshots, and shared prompts.
Review `.mcp_data` and generated restore-point data before sharing workspaces or artifacts.
✓Prompts, selected source files, directory context, code snippets, local paths, command output, change-mode chunks, and brainstorming context may be sent to Gemini CLI and the configured Gemini service.
Avoid referencing `.env` files, credentials, private keys, customer data, unreleased product plans, or regulated datasets in `@` references.
The MCP server and Gemini CLI may log execution details locally; protect terminal history, MCP client logs, and Gemini CLI configuration files.
Google account, project, retention, and telemetry behavior depends on the user's Gemini CLI configuration and Google service terms.
Prerequisites
npm for the `lean-ctx-bin` package, or another upstream-supported install path such as release binaries, Homebrew, Cargo, or AUR.
A local repository workspace where LeanCTX is allowed to read files, run configured shell commands, and store session metadata.
Review of generated MCP client config, shell hooks, per-project `.lean-ctx.toml`, and global `~/.config/lean-ctx/config.toml`.
Explicit allow-listing for any paths outside the current project root that LeanCTX should be permitted to access.
A local repository or workspace you are authorized to index.
Platform-compatible Codebase Memory MCP release binary verified before use.
Agent configuration write access reviewed before running the installer.
Sensitive paths, generated files, vendored code, secrets, and private repositories reviewed before indexing.
Node.js and npx available to the MCP client runtime.
A repository or workspace you are authorized to index and expose to an MCP client.
Ollama embedding model available, or an approved OpenAI-compatible embedding provider and API key.
Static analysis tools installed if you want Context+ to run native linters or compilers.
Node.js 16 or newer.
Google Gemini CLI installed, authenticated, and configured before starting the MCP server.
A Claude Code, Claude Desktop, or compatible MCP client configuration that can run stdio servers.
Review of any `@file` or `@directory` references before sending prompts through Gemini CLI.