Quick Verdict
Context Mode is a local MCP context-management tool for AI coding agents and currently requires Node.js 22.5 or later. It keeps verbose tool output in a local execution and indexing layer, returning selected results to the model so logs, browser snapshots, and large JSON objects consume less context. The code uses Elastic License 2.0 (ELv2), which is source-available and should not be described as MIT, Apache, or OSI open source.
The local core states that it collects no telemetry. That reduces one outbound path but does not change the policies of the host coding agent, selected model, or visited service. Its sandbox is a controlled local subprocess and workspace, not a virtual machine or strong isolation boundary. Optional Insight costs $20 per seat per month and retains structural metadata for 90 days. Use the local core to control output volume, not to execute untrusted code safely.
Best For
- Users of Claude Code, OpenCode, or other MCP clients who process large logs and browser snapshots.
- Developers who want local indexing and selective extraction before model calls.
- Teams able to run Node.js 22.5+ consistently across workstations and CI.
- Organizations willing to approve Insight’s uploaded fields and 90-day retention before enabling it.
- Not suitable where untrusted execution requires VM-grade isolation or ELv2 conflicts with the intended distribution model.
Key Features
- Tool-output compression: processes large logs, pages, repository data, and API output locally.
- MCP integration: exposes indexing, search, and execution to compatible coding clients.
- Session continuity: uses local indexes and pre-compaction snapshots to recover key events.
- Code-based analysis: lets the agent write small scripts instead of reading every raw record.
- No-telemetry local core: does not send product telemetry according to the current project claim.
- Insight analytics: uploads structural metadata for team analysis at $20/seat/month with 90-day retention.
- Local sandbox: constrains workflow execution but remains a host process rather than a VM.
Use Cases
For browser automation, a page snapshot can be much larger than the relevant button and error message. Context Mode can keep the full snapshot local and extract matching fragments. During CI investigation, it can index a complete log and search for the failure chain. Compared with claude-mem, it focuses more on current-context control and event continuity than long-term semantic memory.
Do not rely on the sandbox label for third-party scripts or untrusted repositories. Use a container, VM, restricted account, or isolated CI runner with network, filesystem, and credential controls. Context Mode is context engineering, not malware isolation.
Pricing
The local core is free to use under ELv2. ELv2 generally distinguishes internal use from offering the software as a managed service and includes license-control restrictions. Any plan to embed it in a commercial product or provide a competing hosted service needs review against the actual license text.
Insight currently costs $20 per seat per month and retains structural metadata for 90 days. Before purchase, verify seat definitions, billing, field inventory, data region, deletion, export, admin access, and offboarding. The no-telemetry promise for the local core does not apply to metadata deliberately sent to Insight.
Pros
- Directly addresses coding-agent context inflation.
- Local core has a clear no-product-telemetry claim.
- MCP form supports several compatible clients.
- Selective extraction can avoid repeatedly transmitting full logs.
- Local and Insight data boundaries can be evaluated separately.
Cons
- ELv2 is source-available, not a permissive open-source license.
- Node.js 22.5+ may exceed some enterprise baselines.
- The sandbox is not a VM and cannot contain hostile code.
- Compression and retrieval can omit decisive details.
- Insight uploads structural metadata and keeps it for 90 days.
- Host-agent and model-provider data rules remain separate.
Alternatives
| Tool | Best for | Advantage | Limitation |
|---|---|---|---|
| claude-mem | Claude Code users needing cross-session memory | Stronger automatic memory and semantic recall | More complex memory and provider governance |
| Claude Code | Terminal developers minimizing extra layers | Simpler native toolchain | Large outputs can still consume context quickly |
| OpenCode | Users wanting an open client and model choice | Flexible terminal workflow | Context optimization requires configuration |
| Cursor | Developers preferring an integrated AI IDE | Mature editing, indexing, and agent workflow | Not a standalone local MCP compression layer |
FAQ
Is Context Mode open source?
Source-available is more accurate. It uses Elastic License 2.0 rather than MIT, Apache-2.0, or another OSI-approved license.
Which Node.js version is required?
The current requirement is Node.js 22.5 or later. Check the workstation, editor host, and CI image.
Does the local version send telemetry?
The project states that the local core sends no telemetry. Retrieved summaries still go to the selected model, and external tools retain their own data paths.
Can its sandbox safely run untrusted code?
No. It is not a VM or kernel security boundary. Use containers, VMs, or restricted CI for high-risk execution.
What do Insight and its retention cost?
Insight currently costs $20 per seat per month, and structural metadata is retained for 90 days.
Can ELv2 be used in a commercial project?
Internal commercial use and providing the software as a hosted service are different cases. Review the license text against the actual architecture.
Bottom Line
Context Mode turns local filtering and selective context into an MCP tool and clearly states that its core has no telemetry. Its limits are equally important: Node.js 22.5+, ELv2, a non-VM sandbox, and Insight’s $20 seat price with 90-day metadata retention. Use it for context compression while governing malicious-code isolation, model privacy, and team analytics as separate concerns.