io.github.aaronmsoto/tokenomics-overview-mcp

Applies the cache explainer's published formulas to YOUR token counts. Cache hit rate = cache read / (cache read + cache write + uncached input). Cache cost efficiency = 1 − actual prompt cost / uncached equivalent cost, where uncached equivalent cost = (cache read + cache write + uncached input) × base input price. Give token counts; add `base_input_price_per_mtok` plus either `actual_prompt_cost` or the read/write multipliers (e.g. 0.1 and 1.25) to also get cost efficiency. Compare per workload, not blended.

Glossary lookup across the Foundation's definition page and the cache explainer (e.g. 'KV cache', 'prompt cache', 'tokenmaxing', 'Jevons paradox'). Returns every matching definition with its source.

One rung of the Big-T ladder: what puts a workload in it, the published fix, the stack layers whose levers move it, and the levers tagged with it. `class` accepts 'T(n·k)', 'n*k', 'nk', 'multiplicative', or a slug like 't-n-k'.

Big-T notation, 'Big-O for AI consumption': T(n · k · a) = requests × model calls per request × agent depth, and the six-class complexity ladder from T(1) to T(∞), each with its published fix. Use it to classify how a workload's token cost grows before optimizing constants.

Stated links from tokenomics entities to the other two servers: each carries a quote from the Foundation page that names the target, and a finops://framework/… or focus://spec/… URI to read with those servers (null for FOCUS working-draft identifiers). Filter by `entity_type` + `slug` (e.g. metric cache-hit-rate) or by target `server`.

Canonical markdown of one Tokenomics Foundation document, prefixed with its publication status and followed by CC BY attribution. Pass `section` (an id from list_documents) to get one section instead of the whole document — prefer this for long papers and podcast transcripts.

The Foundation's page on what FOCUS 1.5 does for AI cost, one card per capability: in the working draft (e.g. ModelId and other model properties on SKU Price Details, TokenCacheAction / TokenDirection, PrincipalId / CredentialId), in final Member review (SKU Price dataset, AI billing worked examples), and not in 1.5. Filter by `bucket` (done|review|out) or an `identifier` such as 'TokenCacheAction'. Nothing here is in a ratified FOCUS release until 1.5 is ratified; the FOCUS server serves ratified versions.

One layer of the Seven-Layer Stack: description, components, published tooling table, and the levers that sit at it. `layer` accepts 'l5', '5', 'L5', the layer name ('ai access and control'), or its first word ('energy', 'demand'). L5 is the proxy / router / gateway layer.

Full record for one optimization lever, with provenance and status. Look up by slug or name from list_levers.

One metric's published formula, definition, inputs, 'what good looks like' targets, and its cross-links to the FinOps Framework / FOCUS servers. E.g. 'cache-hit-rate', 'cache-cost-efficiency'.

One persona's responsibilities, key decisions, signals and metrics, and stated links to FinOps Framework capabilities.

The cache explainer's dated snapshot of frontier providers' prompt-cache pricing and behavior (write/read charges as multiples of base input, minimum cacheable prefix, lifetime). A snapshot, not a spec — verify against current provider docs. Optional `provider` filter (anthropic, openai, google).

Start here. Data version, entity counts, and every source document with its publication status (Working Draft, Release Candidate, …). No parameters.

Every ingested tokeneconomics.com document with kind (docs/project/paper/insight), publication status, and section ids for get_document. Optional `kind` filter.

The seven layers (L0 Energy … L5 AI access and control, the routing/gateway layer … L6 Demand and entry points) with the published summary: whether each moves the Big-T multiplier, its key metric, and its primary levers. The lower three set what a token costs; the middle three move a workload between Big-T classes; the top layer sets demand.

Optimization levers as the Foundation publishes them: the stack's primary levers per layer (caching, batching, right-sizing, quantization, complexity routing, budgets, agent caps…), Big-T's five levers (routing, serialization, caching, abstraction transparency, governance), architectural approaches with published token reductions, and the 34× worked-example steps. Filter by `layer` (0-6), `bigt_class`, `kind`, or `document`.

The Foundation's reference metrics with their published formulas: Cache Hit Rate, Cache Cost Efficiency and uncached equivalent cost (cache explainer), net benefit and risk expected loss (value classification), cost per token (definition), and AI unit economics (TCA over realized value, a proposal).

The eight core personas (Engineering, FinOps Practitioner, AI End User, Finance, Product, Procurement, Leadership, Capacity) and four allied ones, with the stages (Production, Consumption, Value) each spans. Filter by `stage` or `core`.

The ten outcome categories (direct dollar wins, labor, intangible) with 'estimate as' units, and the five destinations Finance books realized value to. From 'Classifying and Measuring Realized Value from AI'.

Ranked keyword search across layers, Big-T classes, levers, metrics, personas, value categories, glossary terms, FOCUS tracker items, and every document section. Use when you don't know the slug; follow up with the matching get_* tool or get_document(section).