ENKO
Talk to us
Home/MCP publishing

MCP Tool Publishing

Query company data from the AI you already use.

QueryMind Connect turns approved company queries into tools that AI can call. Ask a question in Claude or ChatGPT, and the connected tool retrieves data within your permissions and returns the results the AI needs to answer.

Visit QueryMindExplore the research

QueryMind Connect

Turn an AI request into a real result

What is MCP tool publishing

A common protocol for AI to call external tools.

MCP defines how AI applications connect to external tools and data. A server describes tools and their inputs; a client invokes a capability and receives its result.

Sources: MCP-TOOLS · MCP-ARCHITECTURE

  1. Tool definitionPurpose and required inputs
  2. AI selectsA tool that fits the question
  3. Server executesCheck user permissions
  4. Return resultData and applied conditions

Why it matters

Work in a familiar AI client

Keep the question and its result in a connected AI application.

Clear business capabilities

Describe what each tool retrieves and which inputs it accepts.

Managed execution scope

Expose tools to authenticated users and enforce access in the execution path.

The value of tool design

Connected tools also need to be discoverable.

Anthropic evaluated Tool Search, which loads relevant definitions from a large tool catalog on demand.

Anthropic internal MCP evaluation · Opus 4 in both conditions · large tool libraries · published 2025-11-24.

MCP evaluation with all tools loaded versus on-demand tool discovery Reported accuracy rose from 49% to 74%. How tools are selected matters after they are connected.

Reported accuracy rose from 49% to 74%. How tools are selected matters after they are connected.

Provider-reported MCP evaluation accuracy. The post does not disclose the sample count or full scoring criteria.

View the comparison data
MCP evaluation with all tools loaded versus on-demand tool discovery
ConfigurationMCP evaluation accuracy (%)
Opus 4 · without Tool Search49%
Opus 4 · with Tool Search74%

Anthropic · Advanced tool use · Tool Search · 2025-11-24

This is a provider experiment with Tool Search, not a comparison of generic AI and QueryMind.

Not a QueryMind Connect measurement or an intrinsic MCP gain; it does not imply QueryMind implements Tool Search.

QueryMind Connect publishes queries with clear business meaning.

BLUEURBAN studies understandable tool definitions and manageable execution boundaries. Approved metrics become fixed query tools whose inputs and access are checked before read-only execution.

Research behind the technology

Anthropic Engineering · 2025-11-24

Find the relevant capability in a large catalog

Discovers relevant tools instead of preloading every definition.

Anthropic Engineering · 2025-09-11

Design tools an AI can understand

Evaluates tool naming, descriptions and returned context against realistic tasks.

Research and application at BLUEURBAN

QueryMind Connect: give AI tools for validated queries.

Define the metric and calculation rules in advance, test the query, then publish it as a tool. When AI calls the tool, the system applies the supplied inputs to that definition and runs the query. The LLM does not write new SQL for every call.

Illustrative example

Publish a validated query; let AI supply the inputs.

A question in an AI client

Show last month’s revenue by department.

AI

  • Choose a relevant query from the tools available to the user.
  • Supply the period and department, then answer from the result.

QueryMind Connect

  • Apply inputs to a query definition that was tested in advance.
  • Check permissions, execute read-only and return the data with its conditions.

Two query paths

SQL generation

AI writes SQL in response to the question

Approved tool

Inputs are applied to a validated query definition

Define the query and its inputs
Describe what each tool queries and the inputs it needs so AI can choose the right one. Definitions that fail execution checks are excluded from publication, with the reason recorded.
Provide tools within the user’s permissions
Identify the user and organization before providing available tools. Check the user’s data permissions when running each query as well.
Run the defined query and return its result
Apply the tool’s inputs to the predefined query and run it read-only. Return the data and applied conditions, and state explicitly when the question cannot be answered.
BLUEURBAN engineering blog
Applied at BLUEURBANQueryMind ConnectMCP tool publishingClaude · ChatGPTConnected AI clients

Frequently asked questions

Q1

Is MCP a language model?

MCP is a protocol. The AI client selects a tool and the server executes a defined business capability.

Q2

Does every call generate new SQL?

Published tools apply arguments to frozen metric definitions. Open-ended natural-language SQL generation is a separate path.

Q3

Does connecting grant access to every record?

Tools are exposed within the caller’s organization and access scope. Connecting does not expand data permissions.

Papers and resources

  1. Introducing advanced tool use on the Claude Developer Platform

    Anthropic Engineering · 2025-11-24

  2. Writing effective tools for agents — with agents

    Anthropic Engineering · 2025-09-11

  3. MCP specification: Tools

    MCP specification · 2025-06-18

  4. MCP Architecture overview

    Model Context Protocol

Bring validated capabilities into the AI you use.

Review query functions, inputs and user permissions to define the publication scope.

Discuss AI tool access