One mcp.json block · bearer key
Outbound without leaving Cursor
Drop Getlead into your MCP config and the agent can read a CSV in your repo, verify every address, build the list and launch the campaign. 34 tools, no context switch.
Project or global scope · the same block works in Windsurf, Zed and VS Code
Claude
ChatGPT
Cursor
Gemini+ any MCP clientsignups.csv → campaignShare142 rows read from the file. 24 would have bounced and were dropped:
One mcp.json entry and the agent panel gains every Getlead tool.
Cursor, Windsurf, Zed, Cline and VS Code take the same config.
A read scope makes sending impossible at the server, not just unlikely.
01 · Local data
Your repo already has the list.
Signup exports, churn queries, a migration script that produces target accounts. The agent reads the file, verifies every address and writes back a clean version with statuses.
- No copy-paste between four tools
- Cleaned file committed alongside your work
- Invalid rows flagged, not silently dropped
Done. The cleaned file is written back with a status column:
Getlead02 · Pipeline
Query to campaign in one session.
Run your own SQL or script to produce accounts, push them into a Getlead list, and launch against them without opening a browser tab.
- Works with whatever already produces your target list
- Campaign settings the agent explains before running
- Hourly ceilings you cannot exceed by accident
Queued. Here is what will send and from where:
Getlead03 · Reporting
Numbers written into your docs.
Ask for last week's campaign stats and have the agent drop the summary straight into a markdown file, committed with everything else.
- Structured data instead of screenshots
- Any format your docs need
- Reproducible next week with the same prompt
Written. The weakest steps are called out at the top:
GetleadBest uses
Where it earns its keep
Repo-scoped setup
A growth site that runs its own outreach keeps the MCP config next to the code, shared with the team.
Repeatable runs
Same prompt, same filters, run monthly. The database is live, so each run returns fresh contacts.
No browser needed
Sourcing, verification and sending happen from the agent panel you already have open.
Setup
Connected in under two minutes
Mint a scoped key
Read only for research projects, read and write when the agent should send.
Add one JSON block
Project scope in .cursor/mcp.json, or global scope in Cursor's MCP settings pane.
Reopen the agent panel
Getlead appears with its tools. Ask it to call get_account as a smoke test.
Copy-paste config and every edge case: the setup docs.
Guardrails come with it
Scoped keys, hourly ceilings on campaigns and replies, a full audit log, and lead text returned as untrusted data. An assistant with your mailbox needs limits that do not depend on the assistant behaving.
“My three SDRs share one license instead of three Apollo seats. 31 meetings last month and the data is cleaner.”
“Skeptical about the flat price, honestly. Landed my first client in 11 days and never looked at the competition again.”
“The niche filters are the secret. Tight list of 400 ops leaders, personalized opener, 6.4% reply rate on the first run.”
Pricing
Already inside Pro and Ultra
No MCP add-on, no per-call fee. Tool calls spend the allowances you already pay for.
FAQ
Before you connect
Project config or global config?
Put the server in .cursor/mcp.json when the work belongs to one repo. Put it in the global config when you want Getlead in every workspace. Never commit a real key: reference an environment variable or keep the file untracked.
Does Cursor support OAuth for MCP servers?
Cursor works cleanly with a bearer header, which is what the setup docs use. If your build supports the OAuth flow, point it at the same URL and it will discover the Getlead authorization server automatically.
Can the agent send email while I am not looking?
Only if you let it. Sending tools are declared as write actions so Cursor asks first, and a read-only key makes it impossible at the server. Hourly campaign and reply ceilings apply on top.
Why use an IDE for outbound at all?
Because the data is often already there: a CSV in the repo, a customer list in a migration, a script that produces target accounts. The agent reads it and turns it into a verified, sending list with no copy-paste round trip.
Does this work in other MCP-capable editors?
Yes. Windsurf, Zed, Cline and the VS Code MCP extension take the same URL and header. This is a standard remote MCP server, not a Cursor-specific integration.
Put outbound where you already work
One config block and the agent can research, verify and send.
Mint an API key