Every change Wellknown observed on this MCP server, newest first, with what it was before and what it became. Tool-surface changes carry the definition diff. Nothing here is edited after the fact.
Changed the definition of "auto_route_task_by_quality_tier", "check_context_cliff", "compress_and_fold_context" and 6 more
ComputesSimulates adeterministic6-turn coding workflow cost projection comparing unrouted flagship execution against tiered routing acrossfree creditgrant pools and warm-cachedproductionGA models.Purely read-only, statelessStateless in-memorycalculation;calculationacquiresrequiring nolocks,authenticationmutatesandzeromutatingstate,noandroutermakessettings.zeroScalesexternalper-turnnetworkcostcalls.linearlyReturnswithJSONcontext_tokenswithrelativecustom_router_mutex_checkto(current_model; when has_custom_router_config is true,sets modereportstoaSHADOW_ADVISORY_ONLY_PRESERVING_USER_ROUTERnon-intrusivewithoutshadowoverridingcomparisontheratheruser'sthanrouter;activewhentargetfalse,IDs.returnsReturnsactiveJSONroutingwithtarget)custom_router_mutex_check, six_turn_cost_breakdown,and(1net_saved_usd.coldParameters:writetask_instructionturn(string+prompt),5has_custom_router_configwarm(booleanread turns),context_tokens (number > 0,andtokensnet_saved_usdper(positiveturn),USDcurrent_modelsaved(stringvsmodelunroutedIDbaseline). Use this tool when planning multi-turnagenttask allocation; use check_context_cliffinsteadfor single-turncontextlengthwindowchecksthresholdorchecks.reconcile_usage_receipt for post-hoc invoice auditing.
⟨12 unchanged words⟩ turn (e.g. 80000 to 200000).","minimum":1,"type":"number"},"current_model":{"default" ⟨13 unchanged words⟩ "has_custom_router_config":{"default":true,"description":"IfWhen true, preserves user's existing router and outputs a non-intrusive shadowcostcomparison;ifwhen false, outputs active routingdirectives.targets.","type":"boolean"},"task_instruction":{"default":"Refactor Nextjs edge routes","description":"Coding oranalysisarchitectural task description used to classify complexity tier.", ⟨4 unchanged words⟩
⟨5 unchanged words⟩ :false,"readOnlyHint":true,"title":"Simulate 6-Turn Quality-Tier Task Routing& Cost Savings"}
⟨30 unchanged words⟩ returns the surcharge multiplier and recommended flat-context alternative.Purely read-onlyDeterministic threshold lookup against the embedded 563-entity UMCEregistry;registryperformsrequiringzerononetworkAPII/O.keys. Returns JSON with crosses_cliff_boolean, active_surcharge_multiplier,and(1.0xrecommended_flat_context_model.belowParameters:cliffmodelvs(string
⟨13 unchanged words⟩ preserving 100% of head lines, tail lines,errorcompilermessageserrors, stack traces, and file paths.Purely read-only,Deterministicdeterministicin-memory string transformationin memory;that mutates no filesorandexternalrequiresstate.no authentication. Returns JSON containing line_accounting_reconciliation (where raw_total_lines=equals kept_original_lines_total+plus omitted_repetitive_lines), byte_and_token_reconciliation
EvaluatesflagshipLLMpricingrate-card tiersand optional upstream response headerstoestimatequantify relay downgrade risk,input-only (20.83x) vs full-turn (17.39x)cost ratios, and remaining codingturns. Purelysessionread-only,turns.deterministicDeterministic in-memory calculationagainst therequiringembeddedno563-entityAPIUMCEkeys;rateunknown
Returns the catalog ofRetrieves verified production GA models and eligiblefree/grantzero-costcreditgrant tiers (Google Cloud Tier 1, AWS Activate Bedrock, OpenRouter :free) from the embedded UMCE registry.Purely read-only, closed-world lookup againstDeterministic local snapshotdata;lookup that issues zeroexternalnetwork probes, requires no API keys, and consumes zero user quota. Returns JSON listingeligiblezero-cost workhorse models, flagship reasoning models, and recommendedtierrole assignmentsforbenchmarkedcurrent_model.against current_model (formatted as 'vendor/model-slug', defaulting to 'openai/gpt-6.1-sol'). Use this tool at session startup to inspect
Queries the563-entityembedded UMCEMerklerate-card registry(451 Production GA LLMs, 34 Frontier Benchmarks, and 78 Cloud Infrastructure Services) filteredbyprovidervendornamenamespaceandtoreturnsretrieve canonical model IDs, input/output/cache USDratesprices per 1M tokens, context windows, and Merkle leaf hashes.Purely read-onlyDeterministic localcataloglookupquery;requiringmakesnozero
Computes the exact single-callReordersanda10-callsingleUSDprompt'ssavingsdynamicachievedmid-turnbytokensplacingafter static prefix tokensbefore dynamic mid-prompt tokenstosopreventthe1.25xstaticcold-cacheblockinvalidation
CalculatesProjects 10-turnpromptsession cachingeconomicscosts for asingletarget model under Case A (100% static prefix: 1 cold turn + 9 warm cached turns) and Case B (90% static prefix + 10% dynamic tail) andreturns provider-specific cacheoutputsbreakpointproviderconfigurationcache-control templates.Purely read-only, deterministicIn-memory actuarialcalculation;lookupmodifiesrequiring noliveAPIprompts
⟨19 unchanged words⟩ relay multiplier markups and unutilized prefix cache waste.Purely read-only, deterministicDeterministic mathematicalaudit;auditmakesinzeromemoryexternalrequiringnetworknocalls.authentication. Returns JSON with expected_theoretical_cost_usd, billing_variance_usd, variance_status ('OVERCHARGED_BY_RELAY_OR_MARKUP' when variance > $0.001 else 'EXACT_RATE_CARD_MATCH'), same_model_unutilized_cache_waste_usd, and net_recoverable_savings_usd.Parameters:Enforcesmodelcached_tokens(string),<= prompt_tokens(number >= 0,
Added "auto_route_task_by_quality_tier", "compress_and_fold_context", "detect_model_downgrade_and_substitution" and 4 more; removed "etuki_get_audit_evidence" and "etuki_search_models"; changed the definition of "check_context_cliff" and "list_supported_models" (4 tools before, 9 now)
CheckEvaluatesonewhetherLLMarequest'sprompt tokencountscountagainstcrossesavendorpricinglong-contextsnapshotsurcharge
Removed "chunk_below_context_cliff", "compress_and_fold_context" and "pin_single_model_prefix_cache"; changed the definition of "check_context_cliff", "etuki_get_audit_evidence", "etuki_search_models" and 1 more (7 tools before, 4 now)
CalculatesCheckexactone LLM request's tokencostcountsonagainstANYamodelpricing(nativelysnapshotsupportsandGLM-5v-Turbo,
Recently observed → Live
Added "chunk_below_context_cliff", "compress_and_fold_context" and "pin_single_model_prefix_cache"; changed the definition of "check_context_cliff", "etuki_get_audit_evidence", "etuki_search_models" and 1 more (4 tools before, 7 now)
Check one LLMCalculatesrequest'sexact tokencountscostagainstonaANYpricingmodelsnapshot(nativelyandsupportsreturnGLM-5v-Turbo,
Live → Recently observed
Recently observed → Live
Live → Recently observed
Certificate recorded, valid to 2027-01-14
Authorization not required
Unknown → Live
First tool surface recorded: 4 tools (server version 40.8.6)
https://www.etuki.com/api/v1/mcp (mcp_streamable_http) — from mcp_registry, with the record
Showing the latest 13 events. The API returns up to 500 and filters by kind: ?kind=tool_surface_changed
⟨4 unchanged words⟩ default":"openai/gpt-6.1-sol","description":"Model ID ('vendor/model-slug') tocheckevaluateforagainst context-window pricing cliffs.","type":"string"} ⟨3 unchanged words⟩ ,"description":"Input prompt token count toevaluatetest against vendor cliff thresholds.","minimum":0,"type":"number"}},"type":"object"}
⟨4 unchanged words⟩ ":{"default":"openai/gpt-6.1-sol","description":"ModelCanonical model ID ('vendor/model-slug') used tocalculatepriceUSDeliminatedsavingsinputfromtokenseliminatedintokens.USD.","type":"string"},"raw_text":{ ⟨16 unchanged words⟩
⟨7 unchanged words⟩ :"Cumulative USD quota spent in the currentsession (>=coding0).session.","minimum":0,"type":"number"},"model":{"default":"openai/gpt-6-astra","description":"ModelCanonical provider/model ID to evaluate (e.g. 'openai/gpt-6-astra' or 'openai/gpt-6.1-sol'). ⟨3 unchanged words⟩ default":137.45,"description":"Remaining USD quotabalance (>= 0);balance;addedcombinedtowith burned_usd tocomputederivetotalinitialpool.pool size.","minimum":0,"type":"number"}},"type":"object"}
⟨4 unchanged words⟩ openWorldHint":false,"readOnlyHint":true,"title":"EstimateEvaluate Relay Downgrade Risk &EnduranceSessionRatiosEndurance"}
⟨6 unchanged words⟩ ,"description":"Active primary model ID inthe'vendor/model-slug'IDEformat used as the baseline cost reference.","type":"string"}},"type":"object"}
⟨4 unchanged words⟩ ":{"default":"all","description":"VendorCase-insensitive vendor namespace filter ('all', 'openai', 'anthropic', 'google', ⟨8 unchanged words⟩
⟨7 unchanged words⟩ Output completion token count for the turn.","minimum":0,"type":"number"},"dynamic_mid_tokens":{"default":93029,"description":"Token count of volatile per-turn context thatpreviouslyinvalidatesinvalidateddownstreamthecachecache.breakpoints.","minimum":0,"type":"number"},"static_prefix_tokens":{"default" ⟨3 unchanged words⟩ count of invariant system instructions and repository context.","minimum":0,"type":"number"}},"type":"object"}
⟨6 unchanged words⟩ ","description":"Target LLM model ID ('vendor/model-slug', e.g. 'openai/gpt-6.1-sol' or 'openai/gpt-4.1').","type":" ⟨5 unchanged words⟩ ":"Input prompt token count per turn (e.g.>=1500001024orcacheable200000threshold).","minimum":1024,"type":"number"}},"type":"object"}
⟨5 unchanged words⟩ :0.49,"description":"Actual USD amount deductedbyon the provider or relayfor this call.invoice.","minimum":0,"type":"number"},"cached_tokens":{"default" ⟨11 unchanged words⟩ (0 <= cached_tokens <= prompt_tokens).","minimum":0,"type":"number"},"completion_tokens":{"default":4000,"description":"Output completion tokensbilled (>= 0).billed.","minimum":0,"type":"number"},"model":{"default":"openai/gpt-4o-2024-11-20","description":"Billed model ID in 'vendor/model-slug' format (e.g. 'openai/gpt-4o-2024-11-20' or 'openai/gpt-6.1-sol').","type": ⟨4 unchanged words⟩ "description":"Total input prompt tokens billed (must be >= cached_tokens).","minimum":0,"type":"number"}},"type":"object"}
{"additionalProperties":false,"properties":{"allow_async_batch":{"default":true,"type":"boolean"},"budget_gate_usd":{"default":0.06,"minimum":0,"type":"number"},"cache_hit_rate":{"default":0.4,"maximum":1,"minimum":0,"type":"number"},"cold_start_cache_writemodel":{"default":false,"type":"boolean"},"completion_tokens":{"maximum":9007199254740991,"minimum":0,"type":"integeropenai/gpt-6.1-sol"},"model":{"description":"Model IDfrom list_supported_modelsto(gpt-6-1-sol,checkclaude-haiku-5-5,forclaude-sonnet-5-5,context-windowclaude-opus-5-5,pricinggpt-6-luna)cliffs.","type":"string"},"pass_rate":{"default":0.96,"exclusiveMinimum":0,"maximum":1,"type":"number"},"prompt_tokens":{"maximum":9007199254740991,"minimum":0,"typedefault":"integer"}150000,"request_id":{"description":"OptionalInputtracepromptIDtoken(1..64countalphanumeric/underscore/hyphentochars).","maxLength":64,"pattern":"^[a-zA-Z0-9_-]{1,64}$","type":"string"},"retry_rate":{"default":0.05,"maximum":1,"minimum":0,"type":"number"},"routed_cache_hit_rate":{"maximum":1,"minimumevaluate against cliff thresholds.":0,"type":"number"}},"required":["model","prompt_tokens","completion_tokens"],"type":"object"}
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true,"title":"Check Prompt Token Count Against 2x-5x Context Cliffs"}
ReturnQueries the5563-entitydeterministicUMCEcliff-engineMerklemodelsregistrysupported(451 Production GA LLMs, 34 Frontier Benchmarks, and 78 Cloud Infrastructure Services) filtered bycheck_context_cliffprovideralongnamewithandtheirreturnsinput/outputcanonicalratesmodel IDs,cacheinput/output/cache USD rates,cliffpertoken1Mthresholdstokens,soft-blockcontextthresholdswindows,fallbackandtargetsMerkle leaf hashes. Purely read-only local catalog query; makes zero external network calls. Parameters: provider (string vendor filter such as 'openai',pricing_snapshot_id'anthropic', 'google', 'deepseek', or 'all'). Use this tool when browsing or filtering the global rate-card catalog by vendor; use discover_and_probe_account_models instead when selecting starter free-credit andMerkleflagshiproot.routing tiers.
{"additionalProperties":false,"properties":{"provider":{"default":"all","description":"Vendor filter ('all', 'openai', 'anthropic', 'google', 'deepseek', 'qwen', 'moonshotai').","type":"string"}},"type":"object"}
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true,"title":"Query 563-Entity UMCE Rate Card Registry by Provider"}
Computes a deterministic 6-turn coding workflow cost projection comparing unrouted flagship execution against tiered routing across free credit pools and warm-cached production GA models. Purely read-only, stateless in-memory calculation; acquires no locks, mutates zero state, and makes zero external network calls. Returns JSON with custom_router_mutex_check (when has_custom_router_config is true, sets mode to SHADOW_ADVISORY_ONLY_PRESERVING_USER_ROUTER without overriding the user's router; when false, returns active routing target), six_turn_cost_breakdown, and net_saved_usd. Parameters: tas…
Compresses verbose terminal or build logs by folding repetitive middle progress lines while preserving 100% of head lines, tail lines, error messages, stack traces, and file paths. Purely read-only, deterministic string transformation in memory; mutates no files or external state. Returns JSON containing line_accounting_reconciliation (raw_total_lines = kept_original_lines_total + omitted_repetitive_lines), byte_and_token_reconciliation, and folded_context_payload. Parameters: raw_text (string log content to fold), model (string model ID used to price saved tokens). Use this tool when tool ou…
Evaluates flagship LLM pricing tiers and optional upstream response headers to estimate relay downgrade risk, input-only (20.83x) vs full-turn (17.39x) cost ratios, and remaining coding turns. Purely read-only, deterministic in-memory calculation against the embedded 563-entity UMCE rate card; performs zero network I/O, mutates zero state, requires no auth, and consumes zero quota. Returns JSON containing three_distinct_ratio_definitions, equal_budget_and_queried_budget_endurance_matrix, and executable_routing_directive. Parameters: model (string, e.g. 'openai/gpt-6-astra'), burned_usd (numbe…
Returns the catalog of verified production GA models and eligible free/grant credit tiers (Google Cloud Tier 1, AWS Activate Bedrock, OpenRouter :free) from the embedded UMCE registry. Purely read-only, closed-world lookup against local snapshot data; issues zero external network probes, requires no API keys, and consumes zero user quota. Returns JSON listing eligible zero-cost workhorse models, flagship reasoning models, and recommended tier assignments for current_model. Use this tool at session startup to inspect available model tiers before calling auto_route_task_by_quality_tier; use lis…
Computes the exact single-call and 10-call USD savings achieved by placing static prefix tokens before dynamic mid-prompt tokens so the static block hits the 0.1x warm cache rate instead of triggering a 1.25x cold cache write rebuild. Purely read-only advisory simulation; does not mutate live prompts or external state. Returns JSON with before_layout_cost_usd (where dynamic_mid_tokens invalidates the cache breakpoint), after_layout_cost_usd (where static_prefix_tokens stays cached), and net_savings_pct. Parameters: static_prefix_tokens (number >= 0), dynamic_mid_tokens (number >= 0), completi…
Calculates 10-turn prompt caching economics for a single model under Case A (100% static prefix: 1 cold turn + 9 warm cached turns) and Case B (90% static prefix + 10% dynamic tail) and returns provider-specific cache breakpoint configuration templates. Purely read-only, deterministic actuarial calculation; modifies no live prompts or external state. Returns JSON with provider_cache_protocol_mechanism, ten_turn_cost_comparison_on_same_model, and two_vector_actuarial_reference_table. Parameters: model (string model ID), prompt_tokens (number > 0, input tokens per turn). Use this tool when keep…
Audits a single API usage receipt against official 1x manufacturer rate cards from the 563-entity UMCE registry to quantify relay multiplier markups and unutilized prefix cache waste. Purely read-only, deterministic mathematical audit; makes zero external network calls. Returns JSON with expected_theoretical_cost_usd, billing_variance_usd, variance_status ('OVERCHARGED_BY_RELAY_OR_MARKUP' when variance > $0.001 else 'EXACT_RATE_CARD_MATCH'), same_model_unutilized_cache_waste_usd, and net_recoverable_savings_usd. Parameters: model (string), prompt_tokens (number >= 0, total input tokens), cach…
Retrieve the cryptographic snapshot metadata (Merkle root, RS256 public key URL, snapshot date, and 12/12 CI self-test reference) for the current pricing snapshot.
Search and filter across all 563 Merkle-committed LLM API and cloud infrastructure entities in the Etuki catalog by keyword.
{"additionalProperties":false,"properties":{"cache_hit_ratioallow_async_batch":{"default":0true,"descriptiontype":"Current prefix cache hit ratio (0.0 to 1.0)boolean"},"budget_gate_usd":{"default":0.06,"minimum":0,"type":"number"},"expected_output_tokenscache_hit_rate":{"default":40000.4,"descriptionmaximum":1,"Expected completion token countminimum":0,"type":"number"},"cold_start_cache_write":{"default":false,"type":"boolean"},"completion_tokens":{"maximum":9007199254740991,"minimum":0,"type":"integer"},"model":{"description":"Any modelModel ID from list_supported_models (e.g.gpt-6-1-sol,zhipu/glm-5v-turboclaude-haiku-5-5,openai/gpt-4.1claude-sonnet-5-5,claude-3-7-sonnetclaude-opus-5-5,gemini-2.5-progpt-6-luna)","type":"string"},"pass_rate":{"default":0.96,"exclusiveMinimum":0,"maximum":1,"type":"number"},"prompt_tokens":{"maximum":9007199254740991,"minimum":0,"type":"integer"},"request_id":{"description":"InputOptionalprompttracetokenIDcount(1..64 alphanumeric/underscore/hyphen chars).","maxLength":64,"pattern":"^[a-zA-Z0-9_-]{1,64}$","type":"string"},"retry_rate":{"default":0.05,"maximum":1,"minimum":0,"type":"number"},"routed_cache_hit_rate":{"maximum":1,"minimum":0,"type":"number"}},"required":["model","prompt_tokens","completion_tokens"],"type":"object"}
before
—after
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}ReturnsRetrieve the cryptographic snapshot metadata (Merkle root, RS256signaturepublicenvelopekey URL, snapshot date, andpricing12/12snapshotCIverificationself-testmetadata.reference) for the current pricing snapshot.
{"additionalProperties":false,"properties":{},"type":"object"}
before
—after
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}SearchesSearch and filter across all 563indexedMerkle-committed LLMmodelsAPI&and cloud infrastructureservices by keyword (e.g.entitiesglm,inqwen,thedeepseek,Etukiclaude,cataloggpt-4.1,bygemini).keyword.
{"additionalProperties":false,"properties":{"query":{"description":"KeywordModeltoorsearchvendor keyword (e.g. gpt-6.1, claude, gemini, deepseek, hunyuan)","type":"string"}},"required":["query"],"type":"object"}
before
—after
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}ReturnsReturnverifiedtheProduction5GAdeterministic cliff-engine models(includingsupportedZhipubyGLM-5v-Turbo,check_context_cliffOpenAIalongGPT-4.1/4o,withClaudetheir3.7input/outputSonnetrates,Gemini 2.5cachePro/Flashrates,DeepSeek-R1/V3,cliffQwen-Max,tokenKimi-k2)thresholds,withsoft-blockexactthresholds,ratefallbackcardstargets,cache discountspricing_snapshot_id, andcliffMerklethresholds.root.
{"additionalProperties":false,"properties":{},"type":"object"}
before
—after
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}SINGLE-MODEL CLIFF SPLITTER: Splits >128K / >200K token documents into sub-cliff windows for your existing single model so you never pay 2x context-cliff surcharges or suffer 120s gateway timeouts.
SINGLE-MODEL CONTEXT & TOOL-OUTPUT COMPRESSOR: Folds verbose logs, duplicate lines, and oversized tool outputs by 65%-92% in-memory so you save money on your CURRENT single model (works with GLM-5v-Turbo, Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, DeepSeek-R1, Qwen-Max) without switching models or changing Base URL.
SINGLE-MODEL PREFIX CACHE REARRANGER: Saves 75%-90% on your CURRENT single model without switching models! Automatically strips volatile timestamps/UUIDs from your system prefix, moves them to the suffix, folds old middle turns, and aligns 4 cache breakpoints.
{"additionalProperties":false,"properties":{"allow_async_batch":{"default":true,"type":"boolean"},"budget_gate_usd":{"default":0.06,"minimum":0,"type":"number"},"cache_hit_rate":{"default":0.4,"maximum":1,"minimum":0,"type":"number"},"cold_start_cache_writecache_hit_ratio":{"default":false,"type":"boolean"},"completion_tokens":{"maximum":9007199254740991,"minimum":0,"type":"integer"},"model":{"description":"Model IDCurrentfromprefixlist_supported_modelscache(gpt-6-1-sol,hitclaude-haiku-5-5,ratioclaude-sonnet-5-5,(0.0claude-opus-5-5,togpt-6-luna1.0)","type":"stringnumber"},"pass_rateexpected_output_tokens":{"default":0.964000,"exclusiveMinimumdescription":0,"maximumExpected completion token count":1,"type":"number"},"prompt_tokens":{"maximum":9007199254740991,"minimum":0,"type":"integer"},"request_idmodel":{"description":"OptionalAnytracemodel ID (1..64 alphanumeric/underscore/hyphene.g.chars)."zhipu/glm-5v-turbo,"maxLength":64openai/gpt-4.1,"pattern":"^[a-zA-Z0-9_-]{1claude-3-7-sonnet,64}$gemini-2.5-pro)","type":"string"},"retry_rate":{"default":0.05,"maximum":1,"minimum":0,"type":"number"},"routed_cache_hit_rateprompt_tokens":{"maximumdescription":1,"minimumInput prompt token count":0,"type":"number"}},"required":["model","prompt_tokens","completion_tokens"],"type":"object"}
before
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}after
—Retrieve theReturns cryptographicsnapshot metadata (Merkle root, RS256public key URL, snapshotsignaturedateenvelope, and12/12 CI self-test reference) for thepricingcurrentsnapshotpricingverificationsnapshot.metadata.
{"additionalProperties":false,"properties":{},"type":"object"}
before
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}after
—Search and filter acrossSearches all 563Merkle-committedindexed LLMAPImodelsand& cloud infrastructureentitiesservicesinbythekeywordEtuki(e.g.catalogglm,byqwen,keyword.deepseek, claude, gpt-4.1, gemini).
{"additionalProperties":false,"properties":{"query":{"description":"Model or vendor keyword (e.g. gpt-6.1, claude, gemini,Keyworddeepseek,tohunyuan)search","type":"string"}},"required":["query"],"type":"object"}
before
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}after
—Return theReturns5verifieddeterministicProductioncliff-engineGA modelssupported(includingbyZhipucheck_context_cliffGLM-5v-Turbo,alongOpenAIwithGPT-4.1/4o,theirClaudeinput/output3.7ratesSonnet,cacheGeminirates,2.5cliffPro/Flash,tokenDeepSeek-R1/V3,thresholdsQwen-Max,soft-blockKimi-k2)thresholds,withfallbackexacttargetsrate cards,pricing_snapshot_idcache discounts, andMerklecliffroot.thresholds.
{"additionalProperties":false,"properties":{},"type":"object"}
before
{"destructiveHint":false,"idempotentHint":true,"openWorldHint":false,"readOnlyHint":true}after
—SINGLE-MODEL CLIFF SPLITTER: Splits >128K / >200K token documents into sub-cliff windows for your existing single model so you never pay 2x context-cliff surcharges or suffer 120s gateway timeouts.
SINGLE-MODEL CONTEXT & TOOL-OUTPUT COMPRESSOR: Folds verbose logs, duplicate lines, and oversized tool outputs by 65%-92% in-memory so you save money on your CURRENT single model (works with GLM-5v-Turbo, Claude 3.7 Sonnet, GPT-4.1, Gemini 2.5 Pro, DeepSeek-R1, Qwen-Max) without switching models or changing Base URL.
SINGLE-MODEL PREFIX CACHE REARRANGER: Saves 75%-90% on your CURRENT single model without switching models! Automatically strips volatile timestamps/UUIDs from your system prefix, moves them to the suffix, folds old middle turns, and aligns 4 cache breakpoints.