Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@tool-disco discover tools for data analysis and report generation"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A tool discovery-oriented MCP (Model Context Protocol) server that matches tools to natural language goals. It integrates with Azure DevOps to discover tools documented in wikis and uses code-graph-rag for code analysis.
Natural Language Tool Discovery: Describe what you want to accomplish, and tool-disco finds matching tools
Azure DevOps Integration: Scans organization wikis for tool documentation with front matter metadata
Code Graph Integration: Ingests repositories via code-graph-rag-mcp for code analysis
Tool Composition: Automatically generates multi-tool workflows
Proxy Testing: Tests tool candidates using lightweight proxies, scripts, and simple models
Related MCP server: Forage MCP Server
npm install
npm run buildAdd to your MCP client configuration:
{
"mcpServers": {
"tool-disco": {
"command": "node",
"args": ["/path/to/tool-disco/dist/index.js"]
}
}
}discover_toolsDiscover tools matching a natural language goal.
{
"goal": "I need to process CSV files and generate reports",
"context": "Working with sales data",
"constraints": ["Must handle large files", "Output as PDF"]
}scan_azure_devopsScan an Azure DevOps organization for repositories with wikis and extract tool metadata.
{
"organization": "https://dev.azure.com/myorg",
"pat": "your-personal-access-token",
"project": "optional-project-name"
}ingest_repoIngest a repository into the code graph for tool discovery.
{
"repoUrl": "https://github.com/org/repo",
"repoName": "my-repo"
}test_compositionTest a tool composition against the discovery goal.
{
"compositionId": "composed-tool1-tool2",
"inputs": { "file": "data.csv" }
}list_candidatesList all registered tool candidates.
get_discovery_statsGet statistics about discovered tools and ingested repositories.
Document tools in Azure DevOps wikis using YAML front matter:
---
toolName: data_processor
description: Processes data files and extracts insights
tags: data,processing,etl
timeout: 5000
implementation: scripts/processor.js
inputs:
file:
type: string
description: Path to the input file
---
# Data Processor
Documentation for the data processor tool... graph TB
subgraph "Tool Disco MCP Server"
ADO[Azure DevOps Integration]
CG[Code Graph Integration]
TD[Tool Discovery]
PT[Proxy Tester]
ADO --> TD
CG --> TD
TD --> PT
end# Install dependencies
npm install
# Build
npm run build
# Run tests
npm test
# Type check
npm run lintDiscover tools that match a natural language goal description. Returns tool candidates and compositions that can accomplish the task.
Inputs: goal (required), context, examples, constraints.
Get statistics about discovered tools and ingested repositories
Inputs: no named parameters.
Ingest a repository into the code graph for tool discovery
Inputs: repoUrl (required), repoName (required).
Execute a planned tool returned by plan_capabilities_for_goal
Inputs: toolId (required), args.
List all registered tool candidates available for discovery
Inputs: no named parameters.
Plan capabilities from discovered tools and code-graph suggestions, returning the planned tool IDs to invoke.
Inputs: goal (required), project, preferredTags.
Scan an Azure DevOps organization for repositories with wikis and extract tool metadata from front matter
Inputs: pat (required), organization (required), project.
Test a tool composition against the discovery goal using proxy implementations
Inputs: compositionId (required), inputs.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 8 tools
Most tools have clear roles, but discover_tools and plan_capabilities_for_goal both address goal-driven tool selection, which could cause misselection. test_composition adds another layer of ambiguity about how it relates to discover_tools.
All tools follow a consistent verb_noun snake_case pattern (e.g., discover_tools, ingest_repo, invoke_planned_tool). The naming convention is uniform and predictable.
8 tools is well within the ideal 3-15 range, covering distinct stages of the discovery workflow without excess. Each tool appears to serve a necessary function.
The set covers discovery, ingestion, planning, invocation, and stats, forming a cohesive workflow. However, it lacks explicit management of registered candidates beyond Azure DevOps scanning, which could be a minor gap.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
MCP server for progressive tool usage at any scale (see https://klavis.ai)
The Remote MCP server acts as a standardized bridge between LLM applications (like Claude, ChatGPT, and Cursor) and external services, enabling AI agents to access external tools and resources. Its primary capability is providing a centralized search tool to discover other MCP servers and their respective tools. Unlike local implementations, it runs remotely with OAuth authentication and permission controls for security.
Search, vet & assemble MCP servers from your agent: verified tools, risk labels, and trust scores.
