Claude Code MCP Tools: 10,000+ Tools via One CLI Command
Claude Code is the best AI coding agent we've used. It runs in your terminal, has full access to your filesystem and shell, and writes real code against real projects. But out of the box, it can only work with what's on your machine — your files, your git repos, your local tools.
We kept hitting the same wall. We'd be deep in a debugging session and need to check a customer's Stripe payments, or pull someone's deal stage from HubSpot, or look at what queries are driving traffic in Google Search Console. Every time, we'd have to stop what we were doing, open a browser, log into a dashboard, click around, copy some data back, and paste it into the conversation. The AI had all the context about our codebase but zero visibility into the services our code actually talks to.
So we fixed it. We built a CLI that gives Claude Code authenticated access to 10,000+ tools across 500+ providers — every major SaaS platform, database, and API — and the AI handles everything. You just ask for what you need in plain English.
Setup: Three Commands, Then You're Done
Claude Code has a terminal. That's all you need.
1. Install the CLI
pip install mcpbundles
2. Authenticate
This connects to your MCPBundles workspace and stores your API key locally (encrypted at rest) so every future session is pre-authenticated:
mcpbundles connect my_workspace
3. Generate the skill file
mcpbundles init
This creates mcpbundles_tools.md in your project directory — a snapshot of every tool available in your workspace. Claude Code reads it automatically and knows the full discovery-to-execution workflow from the first interaction.
That's it. No JSON config files. No per-project setup. No MCP server configuration to maintain. It works across every project directory from this point on.
Just Talk to It
This is the part that surprised us when we first got it working. You don't learn commands. You don't look up parameter names. You don't write shell scripts. You just tell Claude Code what you need, in whatever words make sense to you:
- "Check if customer
acme-corphas any overdue invoices in Stripe" - "Pull last week's signups from the database and cross-reference with HubSpot engagement"
- "What are our top 10 SEO pages by organic clicks this month?"
- "Send a summary email to the team with the results"
- "Check our Sentry error rates and correlate with the last deployment"
- "Look up this customer in Attio and update their deal stage"
- "Query our PostHog analytics for the signup funnel conversion rate"
Claude Code figures out which services to call, what parameters to pass, and how to interpret the results. It writes and executes the CLI commands itself — you never touch them. You ask the question, it gets the answer.
What Services Are Available?
MCPBundles connects Claude Code to 10,000+ tools across 500+ providers — every category a development team touches day-to-day:
| Category | Services | What the AI Can Do |
|---|---|---|
| Payments & Billing | Stripe, Chargebee, ChurnKey, Recurly | Search customers, list invoices, check subscriptions |
| CRM | HubSpot, Attio, Salesforce, Close | Search contacts, update deals, pull pipeline data |
| Databases | PostgreSQL, MySQL, Supabase | Run parameterized queries, explore schemas |
| Email & Communication | Gmail, SendGrid, Campaign Monitor, Fastmail | Send emails, search inbox, manage campaigns |
| Analytics | PostHog, Plausible, Google Analytics, FullStory | Query events, get funnel data, session replays |
| SEO & Search | Google Search Console, Ahrefs | Rankings, indexing status, keyword research |
| Developer Tools | GitHub, Linear, Sentry, LaunchDarkly | Issues, PRs, error tracking, feature flags |
| Cloud & Infrastructure | AWS, Cloudflare, Vercel | DNS, deployments, server management |
| Project Management | Asana, Notion, Monday.com | Tasks, documents, boards |
| Monitoring | Datadog, PagerDuty, FireHydrant | Alerts, incidents, metrics |
| Finance | QuickBooks, Xero, Brex | Invoices, expenses, reconciliation |
| HR & Recruiting | BambooHR, Recruitee, HiBob | Employee data, applicants |
| Document Management | DocuSeal, PandaDoc | Create, send, track documents |
| Video & Media | InVideo, Shutterstock, Mux | Generate content, search assets |
| Security | SocRadar, CrowdStrike, 1Password | Threat intel, vulnerability scanning |
Browse the full catalog at mcpbundles.com/providers or mcpbundles.com/tools.
Every service is pre-authenticated through your workspace. Your team connects services once in the MCPBundles dashboard — OAuth, API keys, whatever the service needs — and from that point on, every Claude Code session gets access automatically. The AI never asks you for a token or a credential. It just works.
How It Actually Works
When you ask Claude Code to do something that involves an external service, here's what happens behind the scenes. You don't need to know any of this to use it — Claude Code handles all of these steps on its own — but it's useful to understand why it works so well.
The AI discovers what's available
Claude Code doesn't start with a hardcoded list of tools. When it needs to interact with a service, it explores what's connected to your workspace at runtime. It might run something like mcpbundles tools --bundle stripe to see what Stripe operations are available, or mcpbundles search "invoice" to find tools across all services that deal with invoices. You never see these commands unless you look at the terminal output — the AI runs them, reads the results, and decides what to call next.
The AI makes individual tool calls
Once it knows what tools exist and what parameters they take, Claude Code constructs and runs shell commands. A HubSpot contact search might look like:
mcpbundles call search-contacts-d4f --bundle hubspot-crm -- query="acme"
Again — Claude Code writes this, runs it, and reads the JSON that comes back. You asked "find the Acme account in HubSpot" and the AI translated that into the right API call with the right parameters.
The AI chains calls across services
This is where it gets powerful. When Claude Code needs data from multiple services — say, matching Stripe customers against HubSpot contacts — it writes a short Python script and runs it through the CLI's exec command:
mcpbundles exec "
customers = await list_customers(bundle='stripe', limit=10)
for c in customers['data']:
contact = await search_contacts(bundle='hubspot-crm', query=c['email'])
print(c['email'], c['created'], len(contact.get('results', [])), 'HubSpot matches')
"
Every tool you've enabled is available as an async Python function inside exec. The AI writes the code, the CLI handles authentication and execution, and you get the results. This is how you go from "cross-reference Stripe with HubSpot" to actual data in one step — the AI wrote a script that calls both APIs, joined the data, and printed the output.
The AI reads the manual first
This is the part that separates this from raw API access. Every bundle has a skill — structured domain knowledge that tells Claude Code how the service works, what the data model looks like, and what the common mistakes are.
Before making its first call to a service, Claude Code fetches the skill. For Stripe, it learns:
Stripe uses Payment Intents for modern card payments. Customers hold
payment methods; Subscriptions link customers to recurring Prices.
All IDs are strings (pi_, cus_, sub_, etc.).
Gotchas:
- Use PaymentIntent for new integrations, not Charges
- Price is immutable — create new Price, update Subscription
- Use idempotency keys for create/upsert to enable safe retries
So when you ask about a customer's payments, it knows to look up PaymentIntents (not the deprecated Charges API), it knows the ID format starts with pi_, and it knows to use idempotency keys if it needs to create anything. It doesn't guess — it reads the manual first, then acts.
Every major service has one of these:
- HubSpot CRM — property names are internal (
firstname, not "First Name"), associations are directional, deals require a pipeline_id - PostgreSQL — use parameterized reads for exploration, wrap destructive statements in transactions, dry-run previews exist for risky updates
- Gmail — persist historyId checkpoints for delta sync, never interchange message vs thread IDs, 403s usually mean missing OAuth scope
- Attio — use attribute
api_slugin upserts (not display name), discover object schema before record operations
The result is that Claude Code makes fewer mistakes on the first try. It's not fumbling through API docs — it already knows the patterns, the naming conventions, and the footguns.
Real Workflow: Debug a Customer Issue
Here's what this looks like end to end. You're in the middle of something and a support ticket comes in:
"Customer john@acme.com reported they were charged twice. Check Stripe for duplicate payments, pull their account from HubSpot, and draft a response email."
You paste that into Claude Code and walk away for two minutes. When you come back, here's what happened:
- It read the Stripe skill, learned the data model and ID formats
- Searched for the customer in Stripe, found their
cus_ID - Listed their recent PaymentIntents, spotted two charges for the same amount within 30 seconds of each other
- Read the HubSpot skill, learned the property naming convention
- Searched HubSpot for the same email, pulled the company name, deal stage, and lifetime value
- Put it all together — wrote a response email with the charge IDs, the amounts, the timestamps, and the customer's account context
Three services. Five minutes. You never opened a single dashboard.
The thing we keep noticing is how much this changes the way you work. You stop thinking "I need to go look something up" and start thinking "I'll just ask." The barrier between "I wonder what's going on with this customer" and having the answer is now one sentence in your terminal.
Real Workflow: SEO Check Across Tools
Here's another one we run almost daily. Quick SEO status check:
"What are our top 10 organic pages by clicks this month? Cross-reference with Ahrefs domain rating and check if any of the top pages have Sentry errors."
Claude Code reads the Google Search Console skill, pulls the top pages, fetches domain metrics from Ahrefs, checks Sentry for errors on those URLs, and gives you a combined report with traffic, authority, and error data in one view.
One sentence, three services, a report you'd normally spend 15 minutes assembling by hand.
This is what "agentic coding" actually looks like in practice — the agent doesn't just write code, it operates across your entire stack. (We cover more workflows like this in Best MCP Servers.)
Claude Code MCP: CLI vs Native Server Config
Claude Code supports MCP natively — you can configure MCPBundles as a remote MCP server in Claude Code's settings. Both approaches work. Here's when to use each:
CLI (mcpbundles) | Native MCP Config | |
|---|---|---|
| Setup | pip install mcpbundles && mcpbundles connect | JSON config in ~/.claude/ |
| Discovery | Runtime — AI explores dynamically | Pre-defined tool list |
| Cross-service | exec lets the AI chain calls in Python | One tool call at a time |
| Portability | Same CLI works in Cursor, Codex, Windsurf | Claude Code only |
| Best for | Multi-tool workflows, exploration | Simple single-service use |
If you're using Claude Code as an agentic coding tool that needs to pull data from multiple services in a single session, the CLI is the better path. If you just need Claude Code to call one or two specific tools occasionally, native MCP config is simpler.
Getting Started
Three commands:
pip install mcpbundles
mcpbundles connect my_workspace
mcpbundles init
The init command generates a skill file that Claude Code reads automatically. From that point on, your AI coding agent has access to every service your team has connected — Stripe, HubSpot, Postgres, Gmail, Ahrefs, Attio, Google Search Console, and hundreds more.
One API key. One install. Every project.
Next Steps
- Browse available tools: mcpbundles.com/tools — search across 10,000+ tools
- See all providers: mcpbundles.com/providers — every connected service
- Learn about MCP: What is MCP? — the protocol that makes this work
- CLI documentation: CLI setup guide — full reference
Frequently Asked Questions
How do I add MCP tools to Claude Code?
Install the MCPBundles CLI with pip install mcpbundles, run mcpbundles connect to authenticate, then mcpbundles init to generate a skill file. Claude Code reads it automatically and can discover and call tools from the first interaction.
What services can Claude Code access with MCP?
Through MCPBundles, Claude Code can access 10,000+ tools across 500+ providers including Stripe, HubSpot, PostgreSQL, Gmail, GitHub, Ahrefs, Google Search Console, Sentry, PostHog, and hundreds of other production services. Browse the full list at mcpbundles.com/providers.
Does Claude Code need API keys for each service?
No. Your team connects services once in the MCPBundles dashboard with OAuth or API keys. The CLI authenticates with a single MCPBundles API key, and every tool call is routed through the platform with the right credentials automatically.
Can Claude Code chain calls across multiple services?
Yes. The mcpbundles exec command lets Claude Code write Python that calls multiple services in sequence — every enabled tool is available as an async function. The AI writes the code, the CLI handles auth and execution.
Does the CLI work with other AI coding agents?
Yes. The same CLI, same commands, same workflow works in Cursor, Claude Code, Codex, Windsurf, or any agent with terminal access. mcpbundles init generates an appropriate skill file for each tool.