Connect
Create one revocable account token and keep it only in MCP environment configuration.
Tetrees MCP 2.2
Create one revocable account token and keep it only in MCP environment configuration.
Quote and run a stateless Pack preview before buying. The full maximum is reserved first, and unused Points return.
Acquire or buy to unlock BYOK, download, saved checkpoints and explicitly approved owner-scoped growth.
Remove unused provider-key fields. A BYOK key is sent only for its selected request and is never persisted. Keep the version suffix so npx cannot reuse a retired cached package.
{
"mcpServers": {
"tetrees-ai": {
"command": "npx",
"args": ["-y", "https://ex.tetrees.ai/pkg/tetrees-mcp.tgz?v=2.2.0"],
"env": {
"TETREES_API_URL": "https://ex.tetrees.ai/api",
"TETREES_TOKEN": "<Account API access token>",
"TETREES_CLIENT_EXTENSIONS_FILE": "</absolute/path/client-extensions.json>",
"OPENAI_API_KEY": "<optional BYOK>",
"ANTHROPIC_API_KEY": "<optional BYOK>",
"ZAI_API_KEY": "<optional BYOK>"
}
}
}
}Browse the complete authenticated tool surface when you need an exact command.
accept_ai_pack_termsAccept the current Pack seller/runtime terms before a maker mutation.
search_ai_packsSearch only Agent AVCP-published Packs.
list_owned_ai_packsList purchased, free-acquired and seller-owned Packs.
acquire_free_ai_packAdd a free Pack to the account without Stripe.
download_ai_packIssue a short-lived, entitlement-checked .taip download.
get_ai_pack_reportRead the buyer-safe, version-bound Agent AVCP report.
list_agent_modelsInspect OpenAI, Claude and Z.AI availability, capabilities and Point weights.
get_ai_pack_runtime_profileDiscover hosted skills, ephemeral-file policy and auditioned client extensions.
quote_agent_runBind a Pack, model, token ceiling, hosted skills and funding mode before execution.
run_ai_packRun a signed Pack before acquisition with Points, or use Points/BYOK plus saved growth after ownership.
prepare_local_ai_pack_runPrepare a consent-preserving plan for selected MCP tools, resources, prompts, local skills and HTTPS APIs.
execute_client_extensionExecute one auditioned method in the local MCP process through a configured HTTPS API, nested MCP server or no-shell local skill.
propose_ai_pack_growthStage one proposal emitted by a successful run; it is not active yet.
accept_ai_pack_growthApprove and gate a delta. Memory can become owner-active; executable changes require a new Pack version.
list_ai_pack_growthList delta status, selectable checkpoints, the account default, usage and hosted quotas without exposing memory contents.
select_ai_pack_growth_versionPin future runs to the signed base Pack or an accepted owner checkpoint after exact confirmation.
create_ai_pack_draftCreate a seller-owned draft with the complete structured buyer guide.
update_ai_pack_draftUpdate mutable listing metadata without overwriting an immutable Pack version.
upload_ai_packUpload a local TAIP/1 asset through owner/product/version-bound private storage.
upload_ai_pack_imageAttach a sanitized JPEG, PNG, WebP or AVIF catalogue image.
get_ai_pack_submission_readinessList every missing Pack, listing and image requirement.
submit_ai_pack_for_auditionSubmit a complete immutable version to Agent AVCP.
quote_ai_pack_auditionBind the current version, digest and Point cost before audition.
run_ai_pack_auditionSpend confirmed Points and run Agent AVCP.
get_ai_pack_private_reportRead owner-only failing gates and actionable repair evidence.
publish_ai_packPublish only the exact signed artifact that passed every mandatory gate.
Execution skills
Bounded provider tools such as cited web search are priced in the quote.
Use up to five supported files. Only filename, size and digest metadata remain.
Add local skills, MCP services or HTTPS APIs. Writes require exact one-call approval.
MCP tutorial
Create a revocable token in Account and add the configuration above.
Search, inspect the runtime profile, then quote and run a Point-funded stateless preview before acquisition.
Call get_ai_pack_runtime_profile, then quote_agent_run. Confirm available balance and canAfford before running.
Call run_ai_pack with declared skills, base checkpoint 0 and explicit request-scoped files. Budget-limited partial output is preserved.
Free-acquire or purchase before BYOK, download, saved checkpoints, growth acceptance or client-owned extensions.
Accept useful memory, list its checkpoints, and select the signed base or a previous owner checkpoint. Executable changes still require a higher Pack version and fresh Agent AVCP.
Local MCP clients may compose separately approved client-owned tools, resources and prompts. Tetrees receives only the explicit attachmentPaths selected for that run; attachment contents remain ephemeral and are not added to the Pack.
get_ai_pack_runtime_profile({ productId: "<product id>" })
quote_agent_run({
productId: "<product id>",
model: "anthropic:claude-haiku-4-5",
fundingMode: "points",
enabledSkills: ["web_search"],
growthVersion: 0,
maxInputTokens: 8000,
maxOutputTokens: 2000
})
run_ai_pack({
productId: "<product id>",
model: "anthropic:claude-haiku-4-5",
fundingMode: "points",
enabledSkills: ["web_search"],
attachmentPaths: ["/explicit/path/evidence.pdf"],
growthVersion: 0,
prompt: "Compare the attachment with current cited sources."
})
list_ai_pack_growth({ productId: "<product id>" })
select_ai_pack_growth_version({
productId: "<product id>",
sequence: 0,
confirmation: "SELECT_GROWTH_CHECKPOINT"
})
prepare_local_ai_pack_run({
productId: "<product id>",
task: "Save the approved memo",
selectedExtensionIds: ["maker.workspace"]
})
execute_client_extension({
productId: "<product id>",
extensionId: "maker.workspace",
methodId: "save_memo",
arguments: { title: "Approved memo", body: "..." },
writeConfirmation: "APPROVE maker.workspace.save_memo"
})Owned intelligence · reversible
卖家的包版本与你的智能检查点彼此独立。保持签名包不变,同时比较、继续成长或回退你审核过的记忆。
序列 0 在没有 Tetrees 托管的所有者记忆时运行已购包,不会删除后续检查点。
将任一已接受检查点设为此账户与包的默认值;Web 与 MCP 共用该选择。
在报价和运行时传入 growthVersion,无需更改默认值即可比较检查点。
托管记忆受包与账户的槽位和字节配额限制;你自己的 MCP 主机状态不会上传或计入。
AI_WORKLOAD_BUSY returns 429 and Retry-After; nothing is charged. Retry after that delay. After any other mutation timeout, inspect the resulting run, checkpoint or report before retrying.