Pengo runs a remote MCP server with setup guides for Claude, Claude Code and Cursor. ChatGPT works in workspaces and plans where custom MCP connectors are enabled, sometimes by an administrator. Getting a useful answer takes a little more thought, which is what this page is for.
What the connector actually is
Six tools over one account, yours. Five read tracked fairs, one fair in full, its booths, its people and customer-facing credit use. The sixth can set or clear a booth's route decision, visit state and note.
It is not a second product. Nothing new is researched over MCP and nothing is bought there. Your agent reads what your account already holds, and every purchase stays a press in the app against a stated price.
Before you connect
Two things, or the tools will answer honestly and unhelpfully.
- 1Write the description of who you sell to, in settings. Nothing scores without it, and a fair with no scores is a booth list your agent has to read in full.
- 2Buy the exhibitor list on one fair. Until something is unlocked,
list_exhibitorshas nothing to return and the answer looks like a broken connector rather than an empty account.
Connect it
The four clients and the exact command or address for each are on the MCP page. They live there rather than here on purpose: two copies of an install instruction drift, and the one that is wrong is always whichever was edited last.
Claude Code uses one claude mcp add command at user scope. Claude and Cursor take the same HTTPS address in their connector settings. ChatGPT uses that address only where its custom-connector controls are available. All four then open a browser for OAuth approval.
Three questions worth asking first
The tools return structured rows, so the useful prompts are the ones that make your agent compare rows rather than fetch one.
- Plan the walk
- "Read the exhibitor list for Hannover Messe, take everything scoring above 70, and group it by hall so I can walk it in one morning." This is the question the product exists for, and it is the one a floor plan cannot answer.
- Argue with a score
- "Show me the five highest-scoring booths and the sentence each score was decided on." The rationale travels with the row, so you can throw out the ones judged on the wrong sentence before you plan around them.
- Find the people, then mark the stands
- "Who is speaking from the companies I scored above 80, and mark those booths as go." The marking is the one write, and it lands in the app where the rest of your route already is.
Reading a thin answer
An empty or short response almost always means one of four things, and they are different problems.
- No fairs at all: the account is new, or discovery has not run. Fairs arrive in your list when you buy something on one or when a discovery pass scores it.
- A fair with no booths: the exhibitor list has not been bought on that fair.
get_eventsays how far the research got. - Booths with no scores: the directory is held and scoring is a separate purchase. The booth list is real; nobody has judged it yet.
- Booths with no
sellsline: those companies had thin pages. They are kept, stamped as looked at, and held back from scoring rather than scored on a name.
What it will not do
- It spends nothing and starts no research.
- It sends nothing to anybody. No email, no message, no post.
- It reads one account. Nothing another customer bought is reachable through it.
- It never edits what the research found. The only write is your own decision about a booth.
- List calls return at most 100 rows at a time and a cursor when more exist.
- The client receives the rows you ask for and may apply its own plan limits, charges and data policies.
If you want the reasoning behind those limits, what Pengo will not guess covers the research side of the same rule.