runtype-admin
Inspect, debug, or manage a live Runtype account through MCP or CLI; mutate only within the requested scope.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- runtype-admin
- description
- Inspect, debug, or manage a live Runtype account through MCP or CLI; mutate only within the requested scope.
- argument-hint
- [live account operation or debugging task]
- user-invocable
- true
02 资源文件 1
- 参考文档agents/openai.yaml521 B
03 SKILL.md 正文
Runtype Admin
Use this skill for live account operations. Every mutation changes the user's Runtype workspace unless you run it against a disposable test account.
Before you start
Runtype has two hosted MCP servers. Both accept browser sign-in on the first call, or a
Runtype API key sent as Authorization: Bearer <key>.
| Server | URL | Tools |
|---|---|---|
| Standard MCP | https://api.runtype.com/v1/mcp/protocol |
One named tool per operation, such as get_me, list_agents, and list_runs |
| Code Mode MCP | https://code.runtype.com/mcp |
search, execute, get_platform_documentation, search_documentation, get_build_instructions, generate_persona_embed_code, get_persona_theme_reference |
To check which server is connected, look at the tool list. Code Mode lists only the seven
tools in the table. Call get_me (on Code Mode, execute with
async (runtype) => runtype.getMe()) to confirm the organization before you change anything.
Operating rules
- Read freely once you have identified the resource ID with a
list_*orget_*call. - Write to a single resource only when the user asked for that change. Follow Update resources safely.
- Use the
runtype-build-productskill for builds that span several resources. - Get explicit confirmation before destructive, bulk, or hard-to-reverse actions. When you
ask, state what else is removed and offer the reversible alternative:
delete_productremoves the product's surfaces, surface items, and surface keys. It keeps the flows and agents.bulk_delete_recordsanddelete_conversationare irreversible.delete_schedulealso deletes the schedule's run history. Usepause_scheduleinstead.delete_messaging_conversationdeletes the transcript. Close the thread withupdate_messaging_conversationinstead.
- Never claim a mutation succeeded unless the tool result confirms it.
- Never show secret values. Use metadata only.
Standard MCP tools
Discover before you act:
- Identity and inventory:
get_me,list_products,list_agents,list_flows,list_tools,list_records,list_collections,list_skills,list_schedules,list_secrets,list_conversations,list_model_configs,list_model_configs_grouped,list_provider_keys. - Debugging:
list_runs,trace_execution,trace_conversation,list_logs,get_log_stats,list_schedule_runs,get_record_results,get_record_step_results,get_record_costs. - Validation:
validate_flow,validate_code,validate_product,validate_product_flow,validate_product_agent,validate_product_surface,validate_product_tool. - Test runs:
execute_agent,run_flow. - Evals and batches:
create_eval_suite,run_eval_suite,add_eval_cases,add_eval_case_from_execution,submit_eval,list_eval_batches,get_eval_results,compare_eval,compare_eval_record,analyze_eval_steps,submit_batch,get_batch_status,get_batch_summary,cancel_batch. - Versions and releases:
list_flow_versions,get_published_flow_version,publish_flow_version,list_agent_versions,get_published_agent_version,publish_agent_version,list_agent_aliases,rollback_agent_alias,list_agent_deployments.
Surfaces belong to a product. List or change them with the product ID in hand.
Use list_model_configs to see which models the account can run. list_available_models
is the platform catalog: its isEnabled and isDefault fields are always null or false,
so it cannot tell you whether a model is enabled.
Large inventories (list_products, list_agents, list_flows, list_tools,
list_records, list_collections, list_skills, and list_schedules) return compact
string previews by default. Use that shape to find the resource, then call the matching
get_* tool. Pass view: "full" only when you need complete strings across the whole
page. list_agents also accepts agent_type (runtype, external, or
claude_managed) when you know the type.
Code Mode MCP
Call search first when you are unsure which method to use:
<!-- prettier-ignore -->
(spec) => spec.categories
Then inspect one method:
<!-- prettier-ignore -->
(spec) => spec.methods.updateAgent
Call execute and shape the result inside the function. Filter, project, and aggregate
before you return data, so large responses do not fill the context:
<!-- prettier-ignore -->
async (runtype) => {
const { data } = await runtype.listProducts({ limit: 20 })
return data.map((p) => ({ id: p.id, name: p.name }))
}
Pass an arrow function with one parameter, such as async (runtype) => .... Do not add a
leading semicolon: the call fails with a syntax or validation error.
Use the CLI
Sign in with runtype auth login, or set RUNTYPE_API_KEY. The CLI calls either MCP
server directly:
runtype mcp tools
runtype mcp call list_runs -F limit=20 --json
runtype mcp call execute --server code -F code=@query.js --json
Use -f key=value for string values and -F key=value for JSON values. For log reads,
runtype logs list and runtype logs stats take the same filters as list_logs, such as
--execution, --agent, --status, and --from.
Debug a failed run
- Find the run with
list_runs. Filter byagent_id,flow_id,conversation_id,surface_id,record_id, orbatch_execution_id, and passstatus: "failed". Subagent and nested-flow runs fold under their parent in the unfiltered list. Passinclude_childrento list them flat.list_runscovers the last 24 hours unless you passfrom. For a schedule, start withlist_schedule_runs. - Pull the trace for that run:
trace_execution, ortrace_conversationfor a multi-turn conversation. - Use
list_logsafter the trace. Filter by execution ID, conversation ID, level, status, or category.list_logsandget_log_statssearch only the last 24 hours unless you passfromorminutes_ago. An empty result says nothing about older runs. - Inspect the referenced flow, agent, surface, tools, and secrets.
- Validate the corrected payload.
- Apply the smallest change. Re-run the same input with
execute_agentorrun_flow, then capture the case withadd_eval_case_from_executionoradd_eval_cases.
Common causes:
- A missing secret. Check with
check_secrets. - A secret referenced in any form other than
{{secret:NAME}}.{{_internal.*}}variables are blocked at runtime. - An unresolved template variable.
- A tool schema that does not match the call.
- A wrong surface behavior setting.
- A model timeout.
To stop a failing schedule, call pause_schedule. run_schedule_now counts toward the
daily execution limit.
Take over a messaging thread
Messaging threads (Slack, iMessage, SMS, email, Telegram) are separate from assistant
conversations. Their IDs start with mconv_, and list_conversations does not return
them.
- Find the thread with
list_messaging_conversations. Filter bysurface_id,status, oragent_mode. - Call
takeover_messaging_conversation. If another operator took it over at the same time, the call returns 409 and names that operator. - Reply with
send_messaging_message. - Call
resume_messaging_conversationto hand the thread back to the agent.
A takeover lapses after the surface's idle timeout (24 hours by default) without an
operator reply. To start a fresh conversation on an SMS or iMessage surface, close the
thread with update_messaging_conversation. Do not delete it.
Update resources safely
- Read the current configuration before you update it.
- Keep fields that the request does not mention.
- Validate before you update when a validator exists.
- Follow pagination cursors on list tools. On
list_runsandlist_logs, stop paging whenpagination.cursorResetis true: the response is page one again, and its cursor loops forever. - Use
get_platform_documentationfor exact contracts, such as schemas, routes, and the tool catalog (topic="platform-catalog"). Usesearch_documentationfor how-to questions. On Code Mode, usesearchto find method signatures. - Use
get_platform_documentation(topic="dashboard-links")when you give the user links. - Use
get_secret_intake_manifest,submit_secret_intake, andcheck_secretsso that credential values never pass through the conversation. - When
get_tool,create_tool, orupdate_toolreturnsconnection.status: "needs_connection"for an OAuth2 MCP tool, give the userconnection.connectUrland ask them to click Connect. Callget_toolagain to confirm"connected". Never ask for OAuth tokens in chat.
Roll back a change
Before you change a flow or agent, note its published version with
get_published_flow_version or get_published_agent_version.
- Flow: find the earlier version with
list_flow_versions, then publish it withpublish_flow_version. - Agent: publish the earlier version with
publish_agent_version. If the agent uses release aliases, callrollback_agent_aliasand check the result withlist_agent_deployments.
A rollback restores configuration only. It does not undo messages sent, records written, or other external effects.
Get explicit confirmation before you:
- Set
on_conflictto overwrite a flow or agent that is managed as code. Publishing rejects these by default. - Move the
livealias. This needs theAGENTS:DEPLOY:LIVEpermission and the alias's current revision asif_match.
Handle permission errors
On Permission denied (HTTP 403), do not retry with guessed IDs. Tell the user which
permission is missing, or request a new key:
- Call
request_api_keywith the narrowestpermissionsthat cover the task and a specificreason. Keep the returnedcodeVerifier. It cannot be recovered. - Wait for a person to approve the request in the dashboard.
- Redeem it with
runtype api-keys claim <request_id> --verifier <codeVerifier>, or withclaim_api_keyusing the defaultdelivery: "secret".
Never print the key into the conversation.
For deeper reference, load the runtype skill if it is installed. Prefer live MCP
documentation over local files.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.