Token Optimizer MCP: context tools, client hooks and benchmark claims
Measure token savings per AI coding agent, optimize context, and share a live local knowledge graph across 16 CLI clients.
At a glance
- What is it?
- Token Optimizer MCP offers context optimization, client integrations and a local knowledge graph for coding agents. Its repository publishes benchmarks and configuration guidance; results depend on workload, provider pricing and client support.
- Who is it for?
- Evaluate the project on representative tasks, recording both model costs and task quality. Check your client's supported hook capabilities, configured mode, privacy and retention settings, and current provider pricing before enabling it.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly JavaScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- ooples/token-optimizer-mcp582 ★ · +60
| Repository | 2026-09-12 | 2026-11-15 |
|---|---|---|
| ooples/token-optimizer-mcp | 522 | 582 |
60 Stars in the last 64 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Client hooks and repeat reads
The README describes tools and hooks that can redirect expensive operations and return differences for repeat reads. Actual behavior depends on the client and configuration: current integration guidance says the enforcing tier defaults to assist, while enforce enables vetoes and off disables hooks. The README describes fail-open and loop-breaking safeguards; verify them for your installed version and client.
A local knowledge graph
The project describes a local knowledge graph for findings and context shared across supported clients. Its token estimates for carrying and re-deriving findings are project estimates, not universal costs. Review the current storage, retention and access settings for your environment. The documentation says telemetry and its upload beacon are opt-in and off by default; inspect the applicable configuration and network behavior before use.
Caching and cost depend on the provider
The project's comparisons model the interaction between prompt caching and compression. Cache write and read pricing, eligibility and expiry vary by provider and model, so the README's example multipliers should not be treated as universal. Compare billed usage and task outcomes using your own provider's current pricing.
How to read the benchmark tables
The repository publishes benchmark scripts, fixtures and tables comparing its optimizer with alternatives. The current README revises earlier results, including a previously reported loss on codebase exploration, and describes changes to both the implementation and token-counting methodology. Compare the same workload, version, tokenizer and retrieval assumptions, and reproduce results before extrapolating.
Compression and retrieval trade-offs
A shorter prompt can move context into a separate file or retrieval step instead of preserving it directly in the model's view. This may reduce transmitted context while adding storage, latency or additional turns. Inspect the benchmark's retention and retrieval measures alongside its reduction figures, and test whether the information needed for your tasks remains available.
Testing session and task outcomes
The repository includes session and task benchmarks as well as compression measurements. Such tests can help assess cost and task quality, but small samples and a single run per task can leave substantial uncertainty. Run repeated trials on representative work, track latency, model usage and output quality, and report the exact versions and pricing used.
Reported savings and attribution
The README reports a sample of net MCP transport tokens avoided and separates some modeled graph effects from the verified headline. These figures describe the project's stated measurement, not a guaranteed reduction in provider charges. Check how the installed version attributes operations and how it counts expansion, retrieval and client identity.
Package contents and installation review
The npm package is @ooples/token-optimizer-mcp. The current repository manifest lists multiple executable entry points and a postinstall script; these can change between releases. Review the exact package version, scripts, permissions and client configuration before installation. Excluding local databases and logs from the published package does not by itself establish runtime retention or privacy behavior.
Editorial conclusion
Evaluate the project on representative tasks, recording both model costs and task quality. Check your client's supported hook capabilities, configured mode, privacy and retention settings, and current provider pricing before enabling it. The repository's benchmark results are project-reported and can change with the code and test setup.
Frequently asked questions
Does token-optimizer-mcp reduce token cost or just compress bytes?
The project aims to reduce context and cost, but savings depend on the workload, provider, model and cache pricing. Its benchmark pricing assumptions are not universal; compare actual invoices and task quality for your setup.
What happens to a large file read once the token-optimizer-mcp plugin is installed?
Hook behavior depends on client support and mode. The current client integration documentation describes assist as the default for its enforcing tier, with enforce available for vetoes and off to disable hooks. The README also describes fail-open and loop-breaking behavior; check the documentation for your version and client.
Which AI coding clients does token-optimizer-mcp track separately?
The package description names 16 CLI clients sharing a live local knowledge graph, and returned context, cost equivalents and net transport avoided are grouped by operation and MCP handshake identity, so Codex, Claude Code, Gemini and other connected clients get their own rows. Older records without an identity stay explicitly unattributed.
Is there a benchmark where another compressor beats token-optimizer-mcp?
Benchmark results change with the implementation and methodology. The current README reports a revised result for codebase exploration and documents earlier figures; inspect the current harness, workload and comparator before drawing conclusions.
How does token-optimizer-mcp back up its own numbers?
The project publishes a benchmark harness, fixtures and checks for its README tables. These are project-reported results, not independent verification; reproduce the relevant test and measure your own workload before relying on the claimed savings.
Official sources
Was this useful?