Back to Details

mcp-assistant

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

MCP Assistant

MCP Assistant

Use MCP Assistant when a task needs tools from connected MCP servers, especially when many possible servers or tools may be available. Prefer MCP Assistant's discovery and routing tools instead of loading every downstream MCP tool directly into the agent context.
MCP Assistant server:
text
https://api.mcp-assistant.in/mcp
User connection page:
text
https://mcp-assistant.in/mcp
当任务需要来自已连接MCP服务器的工具时，尤其是存在大量可用服务器或工具的情况下，请使用MCP Assistant。优先使用MCP Assistant的发现和路由工具，而非将所有下游MCP工具直接加载到Agent上下文当中。
MCP Assistant服务器：
text
https://api.mcp-assistant.in/mcp
用户连接页面：
text
https://mcp-assistant.in/mcp

What MCP Assistant Provides

MCP Assistant提供的功能

MCP Assistant provides access to 100+ MCP servers such as GitHub, Notion, Zapier, Supabase, Exa, DeepWiki, Apify, Context7 and other connected services.
It exposes meta-tools for dynamic MCP discovery and a CodeMode tool that executes programs inside a secure sandbox for programmatic tool calling, workflow execution, and result processing. Use this to avoid expensive LLM tool-calling loops when a task can be handled by discovering the right tools, inspecting only the needed schemas, executing a small program, and returning the final result.
MCP Assistant可访问100+个MCP服务器，包括GitHub、Notion、Zapier、Supabase、Exa、DeepWiki、Apify、Context7及其他已连接服务。
它提供用于动态MCP发现的元工具，以及一个CodeMode工具，可在安全沙箱内执行程序，用于程序化工具调用、工作流执行和结果处理。当任务可通过发现合适工具、仅检查所需Schema、执行小型程序并返回最终结果来完成时，使用此工具可避免昂贵的LLM工具调用循环。

Core Principle

核心原则

Keep MCP tools discoverable, not always loaded.
Do not assume the agent needs every connected tool schema in context. Use MCP Assistant to search for the relevant capability, inspect only the needed schema, then execute the selected tool or workflow through the sandboxed runner.
保持MCP工具可被发现，而非始终加载。
不要假设Agent上下文需要所有已连接工具的Schema。使用MCP Assistant搜索相关功能，仅检查所需Schema，然后通过沙箱运行器执行选定的工具或工作流。

Available MCP Assistant Tools

可用的MCP Assistant工具

codemode_search_mcp_tools

codemode_search_mcp_tools

Search connected MCP servers for relevant tools.
Use this first when the task may require an external service such as GitHub, Notion, Slack, Gmail, Linear, Supabase, Exa, Zapier, cloud providers, databases, or any other connected MCP server.
Search with the user's goal, not just a tool name. For example:
text
find GitHub issues by label
post a message to Slack
search the web with Exa
create a Notion page
query Supabase rows
If no relevant tools are found, tell the user they likely need to connect the relevant MCP server at:
text
https://mcp-assistant.in/mcp
搜索已连接的MCP服务器以查找相关工具。
当任务可能需要外部服务（如GitHub、Notion、Slack、Gmail、Linear、Supabase、Exa、Zapier、云提供商、数据库或任何其他已连接的MCP服务器）时，首先使用此工具。
搜索时需结合用户目标，而非仅使用工具名称。例如：
text
find GitHub issues by label
post a message to Slack
search the web with Exa
create a Notion page
query Supabase rows
如果未找到相关工具，请告知用户他们可能需要在以下地址连接相关MCP服务器：
text
https://mcp-assistant.in/mcp

codemode_get_mcp_tool_schema

codemode_get_mcp_tool_schema

Inspect the exact schema for a selected MCP tool before calling it through 
codemode_run
.
Use this after 
codemode_search_mcp_tools
returns a likely tool and before execution. Read the required parameters, optional parameters, expected result shape, and any server-specific constraints.
Do not guess parameter names for downstream MCP tools. Inspect the schema first.
在通过
codemode_run
调用选定的MCP工具之前，检查其确切Schema。
在
codemode_search_mcp_tools
返回可能的工具之后、执行之前使用此工具。阅读必填参数、可选参数、预期结果格式以及任何服务器特定约束。
不要猜测下游MCP工具的参数名称，请先检查Schema。

codemode_run

codemode_run

Execute downstream MCP tool calls inside MCP Assistant's secure remote workbench.
Use this to call one or more selected MCP tools after inspecting their schemas. Prefer batching or chaining multiple dependent tool calls inside one 
codemode_run
when it avoids unnecessary agent-visible intermediate results.
Good uses:
  • Search GitHub issues, fetch related PRs, summarize the result.
  • Query a database, filter rows, and return only the final answer.
  • Search Exa, rank results, and return the most relevant sources.
  • Create or update records across connected tools when the steps are clear.
  • Transform, sort, deduplicate, or aggregate tool results before returning them to the agent.
Avoid using 
codemode_run
for a long chain when the agent needs to inspect and decide after each step. In that case, run one stage, examine the result, then continue.
在MCP Assistant的安全远程工作台内执行下游MCP工具调用。
在检查选定工具的Schema之后，使用此工具调用一个或多个选定的MCP工具。当可避免Agent可见的不必要中间结果时，优先在单个
codemode_run
内批量或链式调用多个依赖工具。
适用场景：
  • 搜索GitHub问题、获取相关PR并总结结果。
  • 查询数据库、过滤行并仅返回最终答案。
  • 使用Exa搜索网页、对结果排名并返回最相关的来源。
  • 当步骤明确时，在多个已连接工具间创建或更新记录。
  • 在将工具结果返回给Agent之前，对其进行转换、排序、去重或聚合。
当Agent需要在每一步后检查并决策时，避免使用
codemode_run
执行长链。这种情况下，应运行一个阶段、检查结果，然后继续。

Workflow Automation Tools Are Different

工作流自动化工具的区别

Do not confuse saved workflow tools with ad-hoc MCP tool execution.
Use 
codemode_run
when the user wants to execute connected MCP tools now without creating a saved workflow.
Use 
workflow_run
only to run an existing saved workflow by 
workflow_id
. It enqueues the saved workflow for execution. It is not the general-purpose way to call arbitrary MCP tools.
Use 
schedule_upsert
only to create or update a UTC cron schedule for an existing saved workflow. It should be used when the user explicitly wants an action to run on a schedule or at a scheduled time. It is not needed for normal one-off MCP tool calls.
If the user asks to automate something for later, first create or update the saved script workflow, then use 
schedule_upsert
to schedule it. If the user asks to test a saved workflow immediately, use 
workflow_run
.
不要将已保存的工作流工具与临时MCP工具执行混淆。
当用户希望立即执行已连接的MCP工具而无需创建已保存工作流时，使用
codemode_run
。
仅使用
workflow_run
通过
workflow_id
运行现有已保存工作流。它会将已保存工作流加入执行队列，并非调用任意MCP工具的通用方式。
仅使用
schedule_upsert
为现有已保存工作流创建或更新UTC cron调度。仅当用户明确希望操作按调度或在指定时间运行时使用此工具。常规一次性MCP工具调用无需使用此工具。
如果用户要求稍后自动化某项操作，请先创建或更新已保存的脚本工作流，然后使用
schedule_upsert
进行调度。如果用户要求立即测试已保存工作流，请使用
workflow_run
。

Default Workflow

默认工作流

  1. Decide whether the task needs an external MCP capability.
  2. Call 
    codemode_search_mcp_tools
    with a goal-oriented query.
  3. Choose the smallest set of relevant tools from the search result.
  4. For each selected tool, call 
    codemode_get_mcp_tool_schema
    .
  5. Call 
    codemode_run
    to execute the selected tool call or workflow.
  6. Return the final result to the user, not every intermediate object unless it is useful.
  1. 判断任务是否需要外部MCP功能。
  2. 使用面向目标的查询调用
    codemode_search_mcp_tools
    。
  3. 从搜索结果中选择最小范围的相关工具。
  4. 对每个选定工具，调用
    codemode_get_mcp_tool_schema
    。
  5. 调用
    codemode_run
    执行选定的工具调用或工作流。
  6. 将最终结果返回给用户，除非中间对象有用，否则无需返回所有中间对象。

When To Batch Tool Calls

何时批量调用工具

Batch multiple calls inside 
codemode_run
when:
  • The task has three or more dependent tool calls.
  • Intermediate results only exist to feed later steps.
  • The result needs filtering, sorting, grouping, or transformation.
  • The workflow crosses multiple MCP servers.
  • Returning every intermediate result would bloat the agent context.
Do not batch when:
  • The user needs to approve an action before it happens.
  • The agent must inspect an intermediate result to choose the next step.
  • A tool call is destructive or sends messages externally without clear user intent.
  • Required parameters are missing or ambiguous.
  • The user is asking to schedule a saved automation. Use the workflow scheduling tools for that instead.
在以下情况下，将多个调用批量放入
codemode_run
中：
  • 任务包含三个或更多依赖工具调用。
  • 中间结果仅用于为后续步骤提供数据。
  • 结果需要过滤、排序、分组或转换。
  • 工作流跨多个MCP服务器。
  • 返回所有中间结果会导致Agent上下文膨胀。
在以下情况下，请勿批量调用：
  • 用户需要在操作执行前批准。
  • Agent必须检查中间结果以选择下一步操作。
  • 工具调用具有破坏性，或在用户意图不明确的情况下向外部发送消息。
  • 缺少必填参数或参数不明确。
  • 用户要求调度已保存的自动化任务。此时应使用工作流调度工具。

Safety And Approval

安全与审批

Before making external changes, confirm user intent if the action is destructive, public, irreversible, or sends communication to other people.
Examples that usually need confirmation:
  • Sending Slack, Gmail, Discord, or other outbound messages.
  • Creating, deleting, or modifying production data.
  • Closing issues, merging PRs, changing permissions, or deploying services.
  • Running expensive workflows or broad data exports.
Read-only discovery, search, listing, and summarization can usually proceed without extra confirmation.
在进行外部更改之前，如果操作具有破坏性、公开性、不可逆性，或会向其他人发送通信，请确认用户意图。
通常需要确认的示例：
  • 发送Slack、Gmail、Discord或其他出站消息。
  • 创建、删除或修改生产数据。
  • 关闭问题、合并PR、更改权限或部署服务。
  • 运行成本高昂的工作流或大规模数据导出。
只读的发现、搜索、列出和总结操作通常无需额外确认即可进行。

Failure Handling

故障处理

If 
codemode_search_mcp_tools
finds no relevant tools:
text
I could not find a connected MCP tool for this. Please connect the relevant MCP server at https://mcp-assistant.in/mcp, then I can search again.
If a selected tool schema is unclear, inspect another candidate tool or ask the user for the missing required input.
If 
codemode_run
fails, summarize the error, identify whether it was caused by missing auth, missing parameters, unavailable server tools, or sandbox execution failure, then retry only when the fix is clear.
如果
codemode_search_mcp_tools
未找到相关工具：
text
I could not find a connected MCP tool for this. Please connect the relevant MCP server at https://mcp-assistant.in/mcp, then I can search again.
如果选定工具的Schema不清晰，请检查其他候选工具或向用户询问缺失的必填输入。
如果
codemode_run
失败，请总结错误，判断是由缺少授权、缺少参数、服务器工具不可用还是沙箱执行失败导致，仅在修复方案明确时重试。

Example Patterns

示例模式

Find A GitHub Issue And Summarize Related PRs

查找GitHub问题并总结相关PR

  1. Search tools for 
    GitHub search issues and pull requests
    .
  2. Inspect schemas for the issue search and PR lookup tools.
  3. Run a workflow that searches issues, extracts linked PRs, fetches PR details, and returns a short summary.
  1. 搜索工具，关键词为
    GitHub search issues and pull requests
    。
  2. 检查问题搜索和PR查询工具的Schema。
  3. 运行工作流：搜索问题、提取关联PR、获取PR详情并返回简短总结。

Search Web And Return Ranked Sources

搜索网页并返回排名后的来源

  1. Search tools for 
    Exa web search
    .
  2. Inspect the Exa search tool schema.
  3. Run a workflow that searches, filters low-quality results, ranks sources, and returns the best links with summaries.
  1. 搜索工具，关键词为
    Exa web search
    。
  2. 检查Exa搜索工具的Schema。
  3. 运行工作流：搜索网页、过滤低质量结果、对来源排名并返回最佳链接及摘要。

Create A Notion Page From Processed Data

从处理后的数据创建Notion页面

  1. Search tools for the source system and 
    Notion create page
    .
  2. Inspect schemas for both tools.
  3. Run a workflow that fetches source data, transforms it, then creates the Notion page.
  4. Return the created page URL and a concise summary.
  1. 搜索源系统和
    Notion create page
    相关工具。
  2. 检查两个工具的Schema。
  3. 运行工作流：获取源数据、转换数据、然后创建Notion页面。
  4. 返回创建的页面URL和简洁总结。

Agent Guidance

Agent指导原则

Prefer MCP Assistant when it reduces context bloat or tool-selection complexity.
Use direct local tools or CLIs when they are already available, simpler, and do not require MCP server discovery.
Keep the agent context focused on the user task. Let MCP Assistant handle downstream tool discovery, schema inspection, execution, and result processing whenever that is the smaller interface.
当MCP Assistant可减少上下文膨胀或工具选择复杂度时，优先使用它。
当本地工具或CLI已可用、更简单且无需MCP服务器发现时，使用直接的本地工具或CLI。
保持Agent上下文聚焦于用户任务。只要MCP Assistant提供的接口更简洁，就让它处理下游工具发现、Schema检查、执行和结果处理。