Page 4 of 9 · Server reference
tokenomics-overview-mcp
An agentic interface to AI tokenomics guidance from the Tokenomics Foundation: the Seven-Layer Tokenomics Stack, Big-T notation for how token consumption scales, prompt-cache mechanics and the two reference metrics with a calculator, consumption levers including model routing and AI gateways, personas, value classification, and what FOCUS 1.5 does for AI cost. Every answer states its source's publication status, and stated cross-links point into the other two servers.
tokenomics · get_tokenomics_info {} · list-tools · list-prompts · list-resources · list-resource-templates
What it covers
The FinOps Framework and FOCUS describe how to manage technology spend and what the billing data looks like. This server adds the AI-specific layer: what drives token cost, how consumption grows with usage and autonomy, and which levers bring it down. Counts below are from the server's own entry-point call.
tokenomics · get_tokenomics_info {} → structuredContent.counts
Sources and their status
Most Tokenomics Foundation material is still a working draft or release candidate, so the server reports the status on every answer and its instructions tell agents never to present it as a ratified standard. The 14 documents it serves:
| Document | Status as served |
|---|---|
| What Is AI Tokenomics? | Tokenomics Draft Definition v0.5.2 (September 1 2026) |
| The Seven-Layer Tokenomics Stack (explainer and paper) | Release Candidate (projects index label) |
| Big-T Notation (explainer) | Published project explainer (no status label) |
| Big-T Notation Paper | Full paper (projects index label) |
| AI Tokenomics Prompt Cache Explainer | Release Candidate 1 (2026-09-22) |
| AI Tokenomics Personas and Operating Model | Release Candidate 2 (2026-09-04) |
| Classifying and Measuring Realized Value from AI | Release Candidate 2 (2026-09-16) |
| What FOCUS 1.5 does for AI cost, and what it does not | 1.5 ratification 3 Dec 2026 |
| Total Cost of AI (TCA) | Podcast episode 17 (TCA is a proposal, not ratified) |
| Routing Wars: Tokenomics in Action | Podcast episode 16 (survey data described as early, directional) |
| What Tokenomics Is, and What It Isn't | Member-call synthesis (unlisted page) |
| Tokenomics Goes Official (routing layers, per-watt economy) | Podcast episode 15 |
| The State of AI Tokenomics: Top Takeaways | Podcast episode 18 (final State of AI Tokenomics survey figures) |
tokenomics · get_tokenomics_info {} → content[0].text · list_documents {}
The Seven-Layer Stack
The Foundation's organising model: the lower three layers set what a token costs, the middle three move a workload between Big-T classes, and the top layer sets the demand they all serve. list_layers returns the published summary table verbatim:
| Layer | Moves the multiplier? | Key metric | Primary levers |
|---|---|---|---|
| L0 Energy | No, sets the price and supply of power | Energy per token, PUE | Site by delivered power price and capacity; contract supply; shift deferrable work in time |
| L1 Physical infrastructure | No, sets cost per token | Dollars per token, performance per watt | Match accelerator class to workload; adopt newer generations; place by region |
| L2 Infrastructure management | No, sets effective cost per token | Utilization, idle hours | Batch the troughs; autoscale; pool and share GPUs |
| L3 Inference stack | Yes, toward T(1) and T(log n) | Cache hit rate, time to first token, throughput | Right engine; prefix, KV, and semantic cache; disaggregation |
| L4 Model management and selection | Yes, T(n·k) toward T(n) | Cost per task, quality per dollar | Right-size the model; quantize; configure the runtime; adapt |
| L5 AI access and control | Yes, bounds T(n·k·a) and T(∞) | Per-route cost, fan-out and retries | Complexity routing; budgets; agent caps |
| L6 Demand and entry points | Yes, sets the class a workload requires | Request frequency, context volume | Set requirements per use case; trim and reuse context |
Big-T notation is the companion idea: T(n · k · a) = requests × model calls per request × agent depth, with a six-class ladder from T(1) (cached, constant) through T(n) (linear, the healthy default) to T(∞) (unbounded). The “Moves the multiplier?” column says which classes each layer's levers move.
tokenomics · list_layers {} · get_bigt_notation {}
Tool tour
20 tools in the default surface, grouped by job. Names and parameters are verified
against the live tools/list response (source:
docs/mcp-surface.md).
| Tool | Purpose | Key params |
|---|---|---|
| get_tokenomics_info | Start here. Data version, entity counts, and every source document with its publication status. | — (no params) |
| list_documents | Every ingested document with kind (docs / project / paper / insight), status, and the section ids get_document accepts. | kind (optional) |
| get_document | Canonical markdown of one document, prefixed with its status and followed by CC BY attribution. Pass a section to keep responses small. | document (required); section |
| search_tokenomics | Ranked keyword search across layers, Big-T classes, levers, metrics, personas, value categories, glossary, FOCUS cards and every document section. | query (required); entity_types; limit |
| Tool | Purpose | Key params |
|---|---|---|
| list_layers | The seven layers with the published summary (multiplier effect, key metric, primary levers). | — (no params) |
| get_layer | One layer: description, components, published tooling table, and the levers at it. | layer (required): 'l5', '5', a name, or its first word ('energy', 'demand') |
| get_bigt_notation | T(n · k · a), its three variables and the six-class ladder, each class with its published fix. | — (no params) |
| get_bigt_class | One rung of the ladder: what puts a workload there, the fix, the layers that move it, its levers. | class (required): 'T(n·k)', 'nk', 'multiplicative', 't-n-k'… |
| list_levers | Optimization levers as published: the stack's primary levers, Big-T's five levers, architectural approaches with token reductions, worked-example steps. | layer (0–6); bigt_class; kind; document; limit; cursor |
| get_lever | One lever with provenance and source status. | lever (required) |
| Tool | Purpose | Key params |
|---|---|---|
| list_metrics | The reference metrics with published formulas: Cache Hit Rate, Cache Cost Efficiency, uncached equivalent cost, net benefit, risk expected loss, cost per token, AI unit economics. | — (no params) |
| get_metric | One metric's formula, definition, inputs, “what good looks like” targets, and cross-links. | metric (required) |
| calculate_cache_metrics | Applies the cache explainer's formulas to your token counts and interprets the result. Supply multipliers and a base price, or an actual prompt cost. | cache_read_tokens, cache_write_tokens, uncached_input_tokens (required); base_input_price_per_mtok; cache_read_multiplier; cache_write_multiplier; actual_prompt_cost |
| get_provider_cache_snapshot | The explainer's dated snapshot of Anthropic, OpenAI and Google prompt-cache pricing and behavior. A snapshot, not a spec. | provider (optional) |
| Tool | Purpose | Key params |
|---|---|---|
| list_personas | Eight core and four allied personas with the stages (Production, Consumption, Value) each spans. | stage; core |
| get_persona | One persona's responsibilities, decisions, signals and metrics, with stated FinOps Framework links. | persona (required) |
| list_value_categories | The ten outcome categories with “estimate as” units, and the five destinations Finance books realized value to. | group (optional) |
| define_term | Glossary lookup across the definition page and the cache explainer. | term (required) |
| Tool | Purpose | Key params |
|---|---|---|
| get_focus_ai_tracker | One card per FOCUS 1.5 AI-cost capability: in the working draft, in final review, or not in 1.5. Nothing here is ratified until 1.5 is. | bucket: done|review|out; identifier (e.g. 'TokenCacheAction') |
| get_crosslinks | Stated links into the framework and FOCUS servers, each backed by a verbatim quote, with a finops:// or focus:// URI to read there (null for FOCUS working-draft names). | entity_type; slug; server; limit; cursor |
| Tool | Purpose | Key params |
|---|---|---|
| get_crosswalk UNOFFICIAL | 21 curated mappings no Tokenomics Foundation page states (for example its Cache Hit Rate metric → the framework's Cache Hit Rate KPI), each with its reasoning. Only registered when the environment variable is set. | entity_type; slug |
tokenomics · list-tools (default and with FINOPS_MCP_EXPERIMENTAL=1)
Prompts & resources
Prompts (3)
| Prompt | Purpose | Arguments |
|---|---|---|
| optimize_consumption | Walks a workload through Big-T classification, the stack layers and their levers, cache metrics, and the FinOps/FOCUS cross-links — grounded in Tokenomics Foundation guidance. | workload (required) |
| assess_prompt_caching | Checks a prompt layout against the cache explainer's three rules, then measures Cache Hit Rate and Cache Cost Efficiency with the published formulas. | workload (optional) |
| bigt_review | Reviews an architecture with Big-T notation: identifies n, k, and a, places it on the ladder, and proposes bounds (depth caps, budgets, circuit breakers) where it is agent-multiplicative or unbounded. | architecture (required); suspected_class |
tokenomics · list-prompts (verbatim descriptions and arguments)
Resources — tokenomics://overview/…
5 fixed resources plus 6 URI templates (83 concrete resources listed).
| URI | What it is | Type |
|---|---|---|
| tokenomics://overview/start | Start here: sources, their statuses, and how to navigate | fixed |
| tokenomics://overview/bigt | T(n · k · a), its three variables, and the six classes with fixes | fixed |
| tokenomics://overview/glossary | Terms from the definition page and the cache explainer | fixed |
| tokenomics://overview/focus-tracker | What FOCUS 1.5 does (and does not) do for AI cost, by status | fixed |
| tokenomics://overview/meta/manifest | Data version, crawl time, counts, parse warnings, unregistered pages | fixed, JSON |
| tokenomics://overview/documents/{slug} | Canonical markdown of one source document, with its status | template |
| tokenomics://overview/layers/{slug} | One stack layer (l0 … l6) | template |
| tokenomics://overview/bigt/{slug} | One rung of the Big-T ladder | template |
| tokenomics://overview/levers/{slug} | One lever with provenance | template |
| tokenomics://overview/metrics/{slug} | One metric's formula, definition, targets, and cross-links | template |
| tokenomics://overview/personas/{slug} | One persona's responsibilities, signals, and cross-links | template |
tokenomics · list-resources · list-resource-templates
Working with the other two servers
The servers ship separately, so they never call each other. Instead this server hands the agent a URI to read with the sibling server, and only where a Foundation page names the target: each of the 35 stated cross-links carries the quote that proves it.
A FOCUS 1.5 card, for example, links the cached-token properties to the 1.2 column they will live in, and flags the draft names as not yet ratified:
$ get_crosslinks {"entity_type":"focus-item","slug":"fr-2099"}
- focus-item `fr-2099` → focus-working-draft draft-identifier **TokenCacheAction**
(FOCUS working draft, not in any ratified release yet) — evidence: “TokenCacheAction
takes "Uncached", "Read", "Write", or "Other" on prices that meter only tokens
consumed from a request.” (focus-1-5-for-ai)
- focus-item `fr-2099` → focus-working-draft draft-identifier **TokenDirection** …
- focus-item `fr-2099` → focus column **SkuPriceDetails**
(`focus://spec/1.2/columns/skupricedetails`) — ratified 1.2 column; the FOCUS 1.5
work quoted here is not in it yet — evidence: “They are the first FOCUS-defined
SKU Price Details properties with a fixed set of allowed values.”
tokenomics · get_crosslinks {entity_type: "focus-item", slug: "fr-2099"} (second link shortened)
Worked demo: model routing and AI gateways
Routing is where much of the current AI-cost conversation sits. The chain an agent walks: a definition, the layer that owns routing, its levers, then the evidence across every source.
Define it quoted
Source status: Tokenomics Draft Definition v0.5.2 (September 1 2026) · “What Is AI Tokenomics?”
tokenomics · define_term {term: "model routing"}
Find the layer that owns it: L5 AI access and control
L5 is the gateway layer. Its published description:
Its five published components separate the pieces that products often bundle together:
| Component | Covers |
|---|---|
| Access & Admission | Governed AI entry points, authentication & authorization, traffic blocking & kill switches |
| Proxy (Pass-Through) | Transparent pass-through, rate limits, bounded retries, timeouts, fallbacks across providers, unified API surface |
| Router (Selection) | Intent and complexity detection, model and provider selection, routing by data classification, cache, rules & no-model routes |
| Policy & Guardrails | Pre-flight guardrails & compliance, response validation, request, token & budget limits, agent fan-out & tool limits |
| Attribution Capture | Capture source metadata from L6; validate user, app, agent & team; stamp use case & initiative; enforce and propagate metadata |
The same response carries the published tooling table, which names gateway and router options such as LiteLLM, Portkey, Cloudflare AI Gateway, Kong AI Gateway, OpenRouter and Vercel AI Gateway.
tokenomics · get_layer {layer: "l5"}
List the levers at that layer
3 lever(s): - **Complexity routing** (`l5-complexity-routing`, stack-primary-lever, L5, t-n-k-a/t-infinity) - **Budgets** (`l5-budgets`, stack-primary-lever, L5, t-n-k-a/t-infinity) - **Agent caps** (`l5-agent-caps`, stack-primary-lever, L5, t-n-k-a/t-infinity)
All three bound the amplifying classes, T(n·k·a) and T(∞). As the layer text puts it: “the first is an optimization and the second is a containment: a workload with no ceiling has no worst case.”
tokenomics · list_levers {layer: 5}
Search every source for the evidence
Search ranks sections across all 14 documents. The top hit for “model routing gateway” is the Routing Wars episode's key takeaways:
Episode 18 adds the final survey figure, served as document state-of-ai-tokenomics-takeaways: “86% of organizations are running or evaluating model routing.” And the cache explainer's model-switching section warns that a router that switches models mid-session throws away a warm cache, which is the bridge to the AI caching worked example.
tokenomics · search_tokenomics {query: "model routing gateway", limit: 6} · get_document {document: "state-of-ai-tokenomics-takeaways"}
The one-minute version
Start with get_tokenomics_info {} to see what is served and how settled each source is. Use the stack and Big-T tools to decide where to act, calculate_cache_metrics to measure caching on your own numbers, and get_crosslinks to hand off to the framework and FOCUS servers. Quote the status line with every answer: most of this guidance is a draft or release candidate.