tool-name-normalization-mcp-server
Click on "Install 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-name-normalization-mcp-servercall all probe tools and run report_call_log"
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.
tool-name-normalization-mcp-server
A throwaway HTTP MCP server to answer one question empirically:
Does the MCP host (Cursor specifically) invoke a tool by the exact advertised name, or does it munge the name (lowercase / snake-case) first?
This matters because Glean's MCP server dispatches tool calls by exact,
case-sensitive name lookup. PR
#147939 (May 2025) added
server-side lowercasing specifically because Cursor at the time normalized
Glean Search → glean_search client-side and then invoked the munged name,
breaking lookup. We want to know whether that's still true before dropping
case normalization from the server (so built-in / pyagents CAPITAL_CASE names
and external camelCase gateway names can be preserved verbatim).
Findings (2026-06-25)
Validated: current Cursor invokes MCP tool names verbatim — no lowercasing, no
snake-casing, not even for names with spaces or capitals. All 8 probe tools
(including SCREAMING_SNAKE_TOOL, camelCaseTool, and Glean Search)
round-tripped exactly; report_call_log reported 8 calls, 0 mismatches. The
munging that motivated scio#147939
— Cursor rewriting Glean Search → glean_search before invoking — is fixed.
Full table and caveats in TEST_PLAN.md.
Related MCP server: AgentCore MCP Server
Add to Cursor (one-click)
The Render deployment must be live first. Clicking a button opens Cursor with the server config pre-filled; confirm to install, then run the prompts in TEST_PLAN.md.
Stateful (recommended for Cursor):
Stateless:
How it works
The server advertises tools spanning every casing style:
Tool name | Style |
| control (already snake_case) |
| camelCase |
| PascalCase |
| CAPITAL_CASE (pyagents action-name analog) |
| kebab-case |
| mixed |
| the real external-gateway example |
| contains a space — the literal #147939 case |
It uses a catch-all tools/call handler, so even a call whose name matches
no advertised tool is captured (not rejected). Every call echoes the exact
received name; report_call_log dumps the full advertised-vs-received history.
If Cursor calls SCREAMING_SNAKE_TOOL verbatim → no munging, case preservation
is safe. If it arrives as screaming_snake_tool (or camelCaseTool arrives as
camelcasetool / camel_case_tool) → Cursor still munges, and we need
tolerant inbound matching on the server.
Local
mise run start:http # stateful, port 3000
node testing/probe.mjs # smoke test against itDeploy (Render)
render.yaml defines two free web services (-stateless and -stateful).
Connect the GitHub repo as a Render Blueprint; stable URLs:
https://tool-name-normalization-mcp-server-8ghm.onrender.com/mcphttps://tool-name-normalization-mcp-server.onrender.com/mcp
Test with Cursor
Add the deployed URL as an MCP server in Cursor settings.
In chat, ask Cursor to call each tool from this server, then call
report_call_log.Compare
receivedvs the advertised names. Server-side logs ([tool-name-probe] …) in the Render dashboard show the same data.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/rwjblue-glean/tool-name-normalization-mcp-server'
If you have feedback or need assistance with the MCP directory API, please join our Discord server