MCPBundles Hub
MCPBundles Hub
One endpoint. All your tools. Two ways to connect.
Prefer the terminal? pip install mcpbundles
Two Ways to Connect
Use the MCP URL with any client, or install the CLI for terminal-first workflows
MCP URL
Add one URL to Claude, Cursor, ChatGPT, VS Code, or any MCP client. All your enabled MCP servers appear as tools instantly.
https://mcp.mcpbundles.com/hub/- →Works with any MCP-compatible client
- →Browser-based setup via dashboard
- →OAuth authentication built in
CLI
Two commands and your AI coding agent has access to every tool. Auto-generates a SKILL.md so your AI knows how to use them.
- →Claude Code, Cursor, Windsurf, any coding agent
- →Auto-generates SKILL.md with domain knowledge
- →Call tools directly:
mcpbundles call <tool>
How the Hub Works
Tool overload kills AI performance. The Hub solves it — your AI discovers tools on-demand and writes code to orchestrate them.
Discover
AI browses the catalog and your enabled servers first — it picks a server, then loads only the tools it needs. No dumping entire tool catalogs into context.
Code
AI writes Python behind the scenes to chain tools across servers — processing hundreds of records in a single step.
Insight
The code runs behind the scenes. You get the answer — "7 deals need attention" — not 200 raw records dumped in chat.
What Your AI Can Do
The Hub gives your AI workspace superpowers — discover servers, run code across them, manage connections, and publish listings. Service tools (HubSpot, Stripe, etc.) load on demand after your AI picks a server.
Discover2
Browse the catalog
Search every MCP server in the directory
Look up a tool
Get the schema and docs for any tool before calling it
Code Execution1
Chain tools with Python
Your AI writes and runs code to combine HubSpot, Stripe, Postgres, and the rest — hundreds of records in one step
A2A Agents7
List agents
See agents configured in your workspace
View run history
Inspect past agent runs and outcomes
Create or update an agent
Configure name, instructions, and triggers
Attach a skill
Give an agent domain knowledge from a server
Remove an agent
Delete an agent you no longer need
Remove a skill
Detach a skill from an agent
Run an agent
Kick off a run on demand
Connect & Operate7
See enabled servers
List MCP servers turned on in your workspace and what they can do
Add or connect a server
Paste a remote URL, register a REST API, or enable a catalog server
Turn off a server
Pause a server but keep saved credentials and settings
Remove from workspace
Drop a shared catalog server from the team workspace
Delete a private server
Permanently remove a server you imported or created
Test a server
Run validation checks on configuration and credentials
Check Hub status
Confirm the endpoint is up and responding
Workspace Management5
View tool run history
Audit what your AI called and when
List saved connections
See API keys and OAuth tokens on file
Add or update a connection
Save a new key or refresh an OAuth token
Remove a connection
Delete stored credentials you no longer need
Check plan usage
See workspace limits and current consumption
Publish8
Publish to the directory
List a server you run so others can find it
Prove ownership
Verify domain or email control before publishing
Edit listing details
Update name, description, and skill content
Refresh server metadata
Sync tools, auth, and transport from the live server
Score listing quality
Run checks on tools and skill content
Claim a listing
Take over a published server you own
Archive or restore
Hide a listing temporarily or bring it back
Delete a listing
Permanently remove a published server
The CLI Skill
When you run mcpbundles init, it writes a SKILL.md into your project. Your AI agent immediately knows how to discover and use tools across the marketplace — no prompt engineering required.
Discovery workflow, argument formats, rules for tool names with hash suffixes
Key=value, typed JSON with :=, from file, from stdin
Reusable Python script templates for discovery, chaining, and bulk operations
MCPBundles CLI SkillSKILL.md
Auto-generated by mcpbundles init — teaches your AI the discovery-first workflow for real tools across the marketplace.
MCPBundles CLI
Tools are organized into MCP servers (e.g. stripe, hubspot-crm). mcpbundles tools shows hub meta-tools only. Use --server <slug> for service tools.
Workflow
Arguments
For complex/nested JSON, use exec -f to avoid shell quoting:
Rules
- Tool names have hash suffixes — always discover first, never guess
- server= required in exec — await tool(server="slug", ...) for MCP server tools; platform tools (health_check, discover_mcp_servers) omit it
- Variables don't persist between exec calls — put everything in one script
- Multiple connections — use --as <name> on every command
REST API Providers
Register any REST API with connect_mcp_server using base_url instead of server_url. Creates provider + MCP server + credential with one api_request tool. Auth is auto-injected.
auth_type: bearer (default), header, query_param, basic, none.
What the Hub Makes Possible
Cross-Service Analysis
"Which deals have gone cold and have support issues?" AI checks HubSpot deals, cross-references your ticketing system, and tells you exactly which 7 need attention.
Content Performance
"What topics should I write about?" AI pulls traffic data, identifies top performers, checks keyword opportunities in Ahrefs, and gives specific topic recommendations.
Customer Health Scoring
"Flag accounts with declining engagement." AI identifies top accounts by revenue, pulls engagement metrics, checks support history, and ranks by risk.
CLI-Driven Automation
"Process all unresolved support tickets." Your coding agent writes a script that fetches tickets, categorizes them, updates the CRM, and Slacks the team — all from your terminal.
Focused MCP Servers vs The Hub
Use focused MCP servers for everyday tasks, switch to the Hub when you need to combine data across services.
Focused MCP Servers
Connect a specific MCP server to your client. AI sees only the 5-15 tools it needs — fast, reliable, no confusion.
- → Best for daily tasks in one domain
- → Each MCP server includes a SKILL.md
- → Works with any MCP client
The Hub
One endpoint, all your MCP servers. AI discovers tools on-demand and writes code to chain them — returns insights, not data dumps.
- → Best for cross-service analysis
- → AI writes code to chain tools together
- → Available via MCP URL or CLI