OptimToken: AI and Compute Economics MCP server
io.github.OptimNow/ai-pricing-hubOptimNowstreamable-httpv0.12.1Scanned
For FinOps: what an AI workload costs, where to buy it, when to self-host, what moved. All sourced.
8 tools: 4 change things and 4 only read.
- Reads private data · 1
- Reads text others write · 8
- Can send data out · 0
- 2 findings
Tools listed without sign-in. Remote posture
With Anlyon
4 tools here change things. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the OptimToken: AI and Compute Economics MCP server can do
The OptimToken: AI and Compute Economics MCP server is published by the GitHub account OptimNow. It has 8 tools. It can create or change data (4 tools: estimate-llm-cost, recommend-llm-model, estimate-self-hosting-breakeven and track-llm-price-changes), but cannot delete data. 4 tools only read.
1 tool reads private data (compare-llm-models); 8 tools bring in text other people can write (compare-llm-models, estimate-llm-cost, compare-models-side-by-side and 5 more).
The fixed checks matched 2 findings (tool-shadowing). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed 4 times since Anlyon Radar first saw them.
Last scanned Nov 21, 2026, version 0.12.1.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
8 tools
Changes things 4Claude Code: ask
estimate-llm-costCost a workload with EXACT numbers the caller supplies: arbitrary token counts per request and any monthly volume, not just the 10k/100k/1m presets the other cost tools use.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Cost a workload with EXACT numbers the caller supplies: arbitrary token counts per request and any monthly volume, not just the 10k/100k/1m presets the other cost tools use. Use this for 'about 800 in and 200 out, 4 million calls a month', or to price one named model across every use-case profile. To compare 2-4 named models like for like at a preset volume, use compare-models-side-by-side instead. Provide a model name to get detailed cost breakdowns, or compare costs across all use case presets. Each figure comes twice: list price, and the optimized price achievable with prompt caching and the batch API. The presets carry their own cache-hit and batch assumptions; a custom token shape carries none unless cacheHitRate and/or batchEligible are passed with it, and each cost entry states which applied in `optimizationBasis`. Cost figures are final as returned: rounded server-side and dated by `provenance`, with the lever behind each saving, or the reason there is none, named in the text.
Annotations the server declares
- title: Estimate LLM Cost
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
estimate-self-hosting-breakevenAPI or own capacity: the monthly volume at which renting GPU instances costs the same as paying a model's API per token.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it API or own capacity: the monthly volume at which renting GPU instances costs the same as paying a model's API per token. An API bill grows with every request, a rented instance costs the same whether it is busy or idle, so below the break-even the API is cheaper and above it the fleet is. Give a model, a workload and either an instance from the compute catalogue (instanceType) or an hourly rate of your own (hourlyRate). The answer costs the fleet at every rate the catalogue lists (on demand, spot, 1- and 3-year commitments) against the API at list price, at the optimized price with caching and batch, and at the cheapest host known, and compares them at your volume. Use it for 'at what volume does self-hosting Llama 3.3 70B on a g6e.12xlarge pay off' or 'is our 2M requests a month enough to rent our own GPUs'. It assumes neither a throughput nor an operations cost. Without outputTokensPerSecond each break-even states the output tokens per second the fleet would have to sustain around the clock; with it, the share of the fleet's capacity that volume uses, and whether it is out of reach. Without monthlyOperationsCost the fleet figures are hardware rent only: engineer time, storage, egress and failover are left out, and the text says so. It does not check that the instance can hold the model, and it reports whether the model's weights are published at all. Figures are final as returned and dated by `provenance` (API prices) and `computeProvenance` (instance rates).
Annotations the server declares
- title: Estimate Self-Hosting Break-Even
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
recommend-llm-modelPick a model.Reads text others writeNo findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Pick a model. Returns a ranked top 3 for one workload under optional constraints, each with a per-constraint satisfied/violated breakdown as the evidence. Use this when the user wants an ANSWER rather than a table — 'what should I use for support tickets under $500 a month'. To browse or filter the whole catalogue instead, use compare-llm-models. Constraints: (monthly budget, minimum ELO, required capability, self-hostability). Returns a top 3 as structured facts — efficiency rank, ELO, list and optimized cost, FinOps flag, volatility, and a per-constraint satisfied/violated breakdown. When nothing satisfies every constraint the query is reported as over-constrained, `recommendations` is empty, and the nearest misses are returned in `nearMisses` instead, each carrying the constraint it failed and flagged as a miss in the text. A model whose maker has announced its retirement is never recommended: it is left out, counted in `retiredExcludedCount`, and the ones that would have ranked are named in `retiredNotRecommended` with the notice. The ranking is computed server-side and is final.
Annotations the server declares
- title: Recommend an LLM Model
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
track-llm-price-changesWhat moved.Reads text others write1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Warn
The description contains "Every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "What moved. Every other tool here quotes today's price; this one say..."
Description, as the server wrote it What moved. Every other tool here quotes today's price; this one says whether it has just changed, from the price archive optimtoken.optimnow.io keeps (its first day is `provenance.trackingStart`). Without `models` it returns the moves of the last 7 days across the catalogue: each model whose quoted price moved by 1% or more, with the price before and after and what the move does to a workload's monthly bill. With `models` (1-4 names) it returns each model's quoted price since tracking began, change by change. Use it for 'did any LLM price change this week', 'which models got cheaper', 'has the price of X moved since the summer', 'is this price stable enough to budget on'. Read the flags: most models are sold by several hosts and one of them is quoted, so a move flagged `hostSwitch` is a change of the host quoted, not a repricing, and `outlierHost` is an outlier listing. The moves with no flag are returned apart, as `priceChanges`; no flag means none was established, which is not proof that the model maker repriced. There is no fallback for this data: when the archive cannot be read the tool returns an error, never an empty list that would read as 'nothing moved'.
Annotations the server declares
- title: Track LLM Price Changes
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
Only reads 4Show 4 read-only tools
compare-compute-pricingCompare cloud compute instance pricing across AWS, Azure, GCP, DigitalOcean, OCI, OVH, and Alibaba.Reads text others writeNo findings
Why it is in this group
- The name "compare-compute-pricing" says "compare": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Compare cloud compute instance pricing across AWS, Azure, GCP, DigitalOcean, OCI, OVH, and Alibaba. Filter by region, provider, vCPUs, memory, category, processor, or use case. Use this for 'the cheapest 8 vCPU / 32 GiB instance in Europe', and for what a commitment saves: each row gives the Linux list price in USD per hour on demand, on spot, and under a 1-year and 3-year commitment (savings plan, Google CUD, reservation) as an effective hourly rate, each with its saving against that row's on-demand rate. A row with no commitment rate says so. Not every price column is live: `provenance.priceTypes` says which of each provider's price columns come from a live API, which are static constants, and which are unavailable, and `provenance.staticPriceColumns` lists the non-live ones outright; a savings-plan or reserved rate that appears there is a static estimate, not a live quote. Prices are returned as the catalogue serves them and dated by `provenance`.
Annotations the server declares
- title: Compare Cloud Compute Pricing
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
compare-llm-modelsBrowse and filter the whole LLM catalogue and get back a ranked table: price, quality (ELO), efficiency and capabilities.Reads private dataReads text others writeNo findings
Why it is in this group
- The name "compare-llm-models" says "compare": a word for reading.
What it can reach
- private-dataThe description's first sentence says "browse" and "table": it reads databases.
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
- untrusted-inputThe description's first sentence says "browse": it reads web pages, which anyone can write.
Description, as the server wrote it Browse and filter the whole LLM catalogue and get back a ranked table: price, quality (ELO), efficiency and capabilities. Use this when the user wants to SEE THE FIELD — 'show me models under $1/1M', 'which providers have vision models', 'list open-weight models above ELO 1300', 'the ten cheapest models for a coding task'. The table is ordered by Arena ELO unless `sortBy` asks for efficiency, cost or price, and `limit` keeps the first rows of that order. For a single PICK under a budget use recommend-llm-model; to weigh 2-4 NAMED models against each other use compare-models-side-by-side. Prices come from optimtoken.optimnow.io where reachable; the response's `provenance` says which tier served them and whether they are vendor-verified. Filter by provider, price tier (category), openness, capability, price range, or minimum ELO score. Optionally enrich with business metrics for a use case. Price tier and openness are independent: a model can be Frontier-priced and open-weight at once. Reports both list-price cost and the optimized cost achievable with prompt caching and the batch API. Prices, costs and scores are final as returned: rounded and ranked server-side, and dated by `provenance`. The text block is a numbered table of the same figures.
Annotations the server declares
- title: Compare LLM Models
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
compare-model-hostsWho sells a model, and at what price.Reads text others write1 finding
Why it is in this group
- The name "compare-model-hosts" says "compare": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Findings
- Warn
The description contains "every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "...dependent host) at different rates, and every other tool here quotes ONE of them. For 1-4 named..."
Description, as the server wrote it Who sells a model, and at what price. Most models are sold by several hosts (the model maker, a cloud platform such as Amazon Bedrock, Azure or Vertex, or an independent host) at different rates, and every other tool here quotes ONE of them. For 1-4 named models this returns how many hosts sell each, the cheapest, the dearest, the model maker's own rate, the spread between them, any promotional listing, and what a workload costs per request and per month at each. Use it for 'where is Llama 3.3 70B cheapest', 'is the price you quoted the maker's own', 'how much does the choice of host matter for this model'. To cost a workload at the quoted price with caching and batch, use estimate-llm-cost. Read the caveats it returns: the cheapest host may serve a quantized build, which is not the same product, or a promotional rate that ends. It also lists every host with its rate, cheapest first, the platforms the model is listed on (Direct API, AWS Bedrock, Azure AI Foundry, Google Vertex, OpenRouter: a hand-kept estimate for the major models) and the AWS regions where Amazon Bedrock lists it. `hostList.available` says whether that full list was read: without it only the cheapest, the dearest, the maker's own and promotional listings are known, and the text says so. A missing platform or region is not known, never 'not sold there'. Per-host prices are list prices read nightly and dated by `provenance.hostPricingAsOf`, and a model with no host data says so rather than reading as sold by one host.
Annotations the server declares
- title: Compare Model Hosts
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
compare-models-side-by-sideCompare 2-4 named LLM models against all 8 use-case profiles at a chosen monthly volume, showing list and optimized cost for each.Reads text others writeNo findings
Why it is in this group
- The name "compare-models-side-by-side" says "compare": a word for reading.
What it can reach
- untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
Description, as the server wrote it Compare 2-4 named LLM models against all 8 use-case profiles at a chosen monthly volume, showing list and optimized cost for each. Use when the user names specific models to weigh against each other, rather than filtering the whole catalogue. It compares only the names given: when the user also asks whether another model would be better or cheaper, call recommend-llm-model as well, which ranks the whole catalogue for the workload. If they also supply their own token counts, or a volume outside 10k/100k/1m, use estimate-llm-cost instead. Every name is resolved against the catalogue and the outcome is reported in `resolution` and in the text: a name that matched nothing, matched several models, or duplicated an earlier pick is stated explicitly, and such a name gets no column rather than a zero cost. Prices and costs are final as returned: rounded server-side and dated by `provenance`.
Annotations the server declares
- title: Compare Models Side by Side
- readOnlyHint: true
- destructiveHint: false
- openWorldHint: true
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged outbound.
- Reads private data1
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write8
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:59 UTC, unchanged since Nov 13, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 4
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- WarnThe whole server
The endpoint served tools/list without credentials, and the list has 4 tools that change things (4 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://optimtoken-mcp.optimnow.io/mcp: tools/list returned 8 tools to an anonymous client, including estimate-llm-cost (write), recommend-llm-model (write), estimate-self-hosting-breakeven (write), track-llm-price-changes (write)
Posture changes
- Nov 13, 2026
- Tools that change things, listed without credentials: 2 to 4
Findings
Matches of fixed checks, with the text they matched.
The description contains "every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "...dependent host) at different rates, and every other tool here quotes ONE of them. For 1-4 named..."
The description contains "Every other tool", wording about other tools or servers.
What this check looks forMatched text
description: "What moved. Every other tool here quotes today's price; this one say..."
History
Changes to its tool definitions and to its remote endpoint, newest first.
Tools1 changed.~1 changedView diff
- Changed:compare-models-side-by-sideOnly readsdescription
Tools1 changed.~1 changedView diff
- Changed:estimate-self-hosting-breakevenChanges thingsdescriptioninputSchema
Tools8 changed.~8 changedView diff
- Changed:compare-compute-pricingOnly readsannotations
- Changed:compare-llm-modelsOnly readsannotations
- Changed:compare-model-hostsOnly readsannotations
- Changed:compare-models-side-by-sideOnly readsannotations
- Changed:estimate-llm-costChanges thingsannotations
- Changed:estimate-self-hosting-breakevenChanges thingsannotations
- Changed:recommend-llm-modelChanges thingsannotations
- Changed:track-llm-price-changesChanges thingsannotations
PostureRemote posture changed. Tools that change things, listed without credentials: 2 -> 4.View diff
- Tools that change things, listed without credentials: 2 to 4
Tools3 added, 4 changed.+3 added~4 changedView diff
- Added:compare-model-hostsOnly reads
- Added:estimate-self-hosting-breakevenChanges things
- Added:track-llm-price-changesChanges things
- Changed:compare-compute-pricingOnly readsdescriptioninputSchema
- Changed:compare-llm-modelsOnly readsdescriptioninputSchema
- Changed:estimate-llm-costChanges thingsdescriptioninputSchema
- Changed:recommend-llm-modelChanges thingsdescription
Install
Pinned to version 0.12.1. Values in angle brackets are yours to fill in.
claude mcp add --transport http ai-pricing-hub https://optimtoken-mcp.optimnow.io/mcpclaude mcp add --transport http ai-pricing-hub https://optimtoken-mcp.optimnow.io/mcpRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__ai-pricing-hub__compare-llm-models",
"mcp__ai-pricing-hub__compare-models-side-by-side",
"mcp__ai-pricing-hub__compare-compute-pricing",
"mcp__ai-pricing-hub__compare-model-hosts"
],
"ask": [
"mcp__ai-pricing-hub__estimate-llm-cost",
"mcp__ai-pricing-hub__recommend-llm-model",
"mcp__ai-pricing-hub__estimate-self-hosting-breakeven",
"mcp__ai-pricing-hub__track-llm-price-changes"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as ai-pricing-hub. Review it before you use it.
Capability tags behind 8 rules
- mcp__ai-pricing-hub__compare-llm-models private-data, untrusted-input
- mcp__ai-pricing-hub__estimate-llm-cost untrusted-input
- mcp__ai-pricing-hub__compare-models-side-by-side untrusted-input
- mcp__ai-pricing-hub__recommend-llm-model untrusted-input
- mcp__ai-pricing-hub__compare-compute-pricing untrusted-input
- mcp__ai-pricing-hub__compare-model-hosts untrusted-input
- mcp__ai-pricing-hub__estimate-self-hosting-breakeven untrusted-input
- mcp__ai-pricing-hub__track-llm-price-changes untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/io.github.OptimNow/ai-pricing-hub/verify?hash=72ac05da438633361cc0cf0143e9095322b2b1ba9708b2cc5592ab7354c94d6a" | jq -e '.changed == false'Pinned hash 72ac05da4386, from Nov 21, 2026, 15:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/io.github.OptimNow/ai-pricing-hub/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/io.github.OptimNow/ai-pricing-hub)Source: MCP Registry · repository. Synced Nov 21, 2026, 15:17 UTC.
OptimToken: AI and Compute Economics MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the OptimToken: AI and Compute Economics MCP server safe?
- No scan can promise that. What its definitions show: 8 tools, 4 of which can change, send or delete something, and 2 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the OptimToken: AI and Compute Economics MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the OptimToken: AI and Compute Economics MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the OptimToken: AI and Compute Economics MCP server have?
- 8 tools: 4 write, 4 read. estimate-llm-cost, recommend-llm-model, estimate-self-hosting-breakeven, track-llm-price-changes, compare-llm-models, compare-models-side-by-side, compare-compute-pricing and compare-model-hosts.
Does the OptimToken: AI and Compute Economics MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the OptimToken: AI and Compute Economics MCP server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the OptimToken: AI and Compute Economics MCP server changed its tools?
- Yes, 4 times since Anlyon Radar started tracking it, most recently on Nov 20, 2026. Every change is on this page and in its Atom feed.