Download Latest Version v1.6.2 source code.zip (803.2 kB)
Home / v1.4.1
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-10-31 11.5 kB
v1.4.1 source code.tar.gz 2026-10-31 469.9 kB
v1.4.1 source code.zip 2026-10-31 538.6 kB
Totals: 3 Items   1.0 MB 0

What's New

Change Details
How each model is priced search_models, recommend_model and get_model_schema show how every model is priced, in the text and in the structured pricing. A price that the run sets when it ends, as for an LLM billed per 1M tokens, gives the rates from the model's price note, and the listed estimate when there is one. A live model shows that it is billed turn by turn during the session, with its rates, and a model billed by run time shows its rate per second of run time. Units read per prompt character, per input and output token and per second of audio in inputAudio, and get_model_schema says what a token or character price counts. A price row without a fixed price says that the run is charged for the time it takes.
Prices by image size Models priced by image size show the price of the first 1024 x 1024 block of the output image and, when the model sets them, of each further block and of each block of an input image, up to 4 blocks per image. get_model_schema explains how blocks are counted and gives each row's rates and an example, and its structured pricing carries the same rules.
Unknown task IDs and tokens get_task, wait_for_task and get_task_price say at once when no task with that ID or token is in this workspace, and suggest checking it or finding the task with list_tasks. A taskid that is not a number, such as a task token passed as the ID, gets the answer that taskid is the numeric Task ID and a Task Token goes in tasktoken. get_task_price passes on any other error from the lookup the way get_task does.
What a workflow estimate leaves out After the estimated cost per run, get_workflow lists what the figure leaves out or counts only in part: steps charged by usage after the run, with their rates; steps without a fixed price, or whose price depends on a setting; steps counted as one unit of a per-unit price, or at a default that an earlier step fills; and the Start inputs that change the price, with their prices and the value the figure uses. The structured estimate returns them as notes.
Answers for team members When the connected project belongs to a team in which you are a member without admin rights, the agent tools say who can act. deploy_agent says before any quote that only a team admin can deploy in the team. manage_agent_billing and apply_agent_skills first check whether you may pay for or change that agent, as the person who deployed it may, and otherwise quote and charge nothing. Hints in get_my_agent, list_my_agents and control_agent name who can act: a team admin or the person who deployed the agent starts it, saves credentials and pays for the plan, and only the person who deployed it connects accounts. Before asking for any credential, set_agent_credentials has the assistant check that you created the connected project and deployed the agent, and get_account shows your team role, in the text and as workspace.role.
Template tasks and instructions get_agent_template with include_tasks: true lists the template's scheduled tasks and custom instructions in its text as well as in its structured output, so clients that show only text see them too. Each entry shows its key, whether it is on or off and its description, and a scheduled task also shows its schedule. A line after the lists says that a task or instruction that needs a credential switches on when that credential is saved.
Price updates in skill quotes When an apply_agent_skills change also moves the plan to today's price for the skills it already has, the quote adds a Price update line with the plan's current monthly price and today's price for those skills, and both monthly credit amounts when they differ. The structured output carries the same text in pricing.note.

Enhancements

  • get_model_schema: checks the price rows against the model's settings. When no setting value selects a priced row, the price reads "No fixed price", and an offered value that no row prices is named after the table and in the structured pricing. It shows the notes of the price rows, as the model's page does, and returns them as pricingNotes. It also prints the table for rows keyed by numbers, such as a duration of 10, and the row with the highest count of an input reads "or more", as in qty: 1 or more.
  • search_models, recommend_model, get_model_schema: show your discount on a model when you have one, as text and as discountPercent; it comes off the part of a charge paid from deposits. estimatedCostUsd is a positive estimate of one run, or null when there is none, as for a model priced by its options, and the output schema describes both fields. The search_models description and the models guide say to read the price in get_model_schema before quoting it or naming the cheapest model.
  • run_model, get_task, wait_for_task, list_tasks, upload_file: output links are plain strings in the output schema, so clients that check results against it accept links whose file names have spaces, non-ASCII letters, square brackets or a % sign.
  • upload_file: takes single files. A .zip archive is refused before anything is sent, with a note to pass its public URL to the model or to upload its files one at a time.
  • run_workflow, wait_for_workflow_run, resume_workflow_run: an output of a step that did not finish is listed once, as not available. A step's cause is left out when it only repeats its status, and a step skipped after a cancelled step says so.
  • get_my_agent: the new entry input takes the key of one custom instruction or scheduled task and returns its whole text, in the text and as entry, and a shortened entry in the lists points to it. A stopped or failed agent with no active plan and no extra credits left points to manage_agent_billing: renew for an expired wallet plan, subscribe otherwise. Custom instructions say whether they are on or off, and the prices line adds that a skill change moves the plan to today's prices for all its skills.
  • get_my_agent, list_my_agents, manage_agent_billing: a card plan whose renewal is managed in the Wiro dashboard reads "paid through" its period end, and agent results carry the new optional subscription.atPeriodEnd: renews, ends or unknown.
  • control_agent, manage_agent_billing: when the agent cannot be read right after an action, control_agent reports the new status as unknown (null in the structured result) and manage_agent_billing shows only the agent line, and both point to get_my_agent. After a paid action without a clear answer, manage_agent_billing points to get_my_agent and the wallet balance in get_account before any new attempt.
  • set_agent_credentials: says that saving a credential turns on the scheduled tasks that need it, and that any the user does not want can be turned off.
  • send_agent_message, get_agent_message, get_agent_conversation: on a team agent with private chats, a member who did not deploy the agent is told when a result handed to the background will not appear for their project, so the assistant tells the user instead of waiting. get_agent_conversation says that such results may not be listed for every member's project, and the talk_to_my_agent prompt says to read them unless the turn says they will not appear. The get_agent_conversation description also says that what the agent writes itself, such as a receptionist's call report, comes back as written.
  • get_agent_template: the chat models line names the default model and gives each model's credits per 1M input and output tokens.
  • get_playbook: the agents guide says a plan's price is set at today's prices when the plan starts, moves to Pro or changes its skills, and that renewals keep that price. The models guide says how to quote a price set when the run ends, billed turn by turn or with no fixed price, and how a discount applies. The agents and billing guides say which calls need a team admin or the person who deployed the agent, and that the organization's owner can also pay for a team agent. The workflow-agent guide says to keep keys, tokens and passwords out of a workflow's inputs and prompts.
  • Server instructions: under 4,000 characters, with the rules first. The rules on quotes and change summaries, long jobs, error results, credentials, usage-priced work and other writes all end within the first 2,048 characters, for clients that pass only the start of the instructions to the model. The error rule also covers a result that says part of the work was done.
  • Library: registerTools makes the server list every tool schema, its own tools included, as JSON Schema 2020-12, whether it is connected yet or not, as createMcpServer does. Some MCP clients accept only that dialect. The new advertiseJsonSchema2020_12(server) and withJsonSchema2020_12(transport) exports do the same when you register single tools from @wiro-ai/wiro-mcp/tools, and calling either again changes nothing.
  • Library: createMcpServer caps one tool call at 10000 array elements and object members, and a larger call gets one short error before its input is validated. The new maxToolInputElements option changes the limit, and Infinity turns the check off.
  • Library: WiroClient.waitForTask throws a TaskPollingError right away when the task is not in the workspace, and WiroClient.uploadFile refuses a .zip archive before sending it. @wiro-ai/wiro-mcp/client exports isTaskNotFound, isTaskId and isTaskIdTooLarge, the checks the task tools use. When an error answer is a web page, the WiroApiError message carries the page's title, or else up to 200 characters of its text, and responseBody keeps the full answer.
  • MCP SDK: the package requires @modelcontextprotocol/sdk 1.32.2 or later, which accepts tool and prompt calls without arguments; library users who also depend on the SDK should update it to match.
  • README: lists the new exports and the maxToolInputElements option, and the second library example imports what it uses, creates its client and adds the model tools to an McpServer you already have.
  • npm listing: the package description covers workflows, Studio and agents, and the keywords add workflows and ai-agents.

Tools

Models: search_models, get_model_schema, recommend_model, explore, run_model, wait_for_task, get_task, list_tasks, get_task_price, cancel_task, kill_task, upload_file

Workflows: list_workflows, get_workflow, create_workflow, update_workflow, clone_workflow, delete_workflow, run_workflow, wait_for_workflow_run, resume_workflow_run, list_workflow_runs, save_workflow_from_agent

Agents: send_agent_message, get_agent_message, get_agent_conversation, cancel_agent_message, delete_agent_session, list_studio_results, list_agent_templates, get_agent_template, list_agent_skills, deploy_agent, list_my_agents, get_my_agent, control_agent, update_agent, apply_agent_skills, set_agent_credentials, connect_agent_oauth, disconnect_agent_oauth, manage_agent_custom_skill, manage_agent_billing, get_agent_activity, delete_agent

Account and guides: get_account, get_playbook, search_docs

Full Changelog: https://github.com/wiroai/Wiro-MCP/compare/v1.4.0...v1.4.1

Source: README.md, updated 2026-10-31