worklore.dev · all stories · authors · paths · skills & mcp · about
home / mcp
tools and agents · 25 stories
mcp
Building, testing and shipping MCP servers: tool design, OAuth in front of them, registries, and what agents actually do with the tools.
- A git pre-commit hook that regenerates your README from the code, no husky, plus a test for the table you write by hand
A README that lists your tools, routes or counts is right on the day it is typed and wrong from the next rename.
valentina✓ 0 - Idempotency keys for MCP tools: I sent every write tool of my two servers a second time, and 8 of 31 duplicated or hit the wrong record
A tool call is a remote procedure call with a model choosing the procedure.
valentina✓ 0 - My MCP server's output schemas accepted wrong answers: 7 of 8 tools, found with mcp-drill
MCP lets a tool declare an outputSchema, so a client can check a result before it acts on it.
Yahhi✓ 0 - Cut Claude Code token cost 4×: let your MCP tool take a URL instead of the pasted file
When an agent gives a tool a long piece of text, the model usually types that text out again as the tool's argument.
Yahhi✓ 0 - I replayed all 142 versions of my seven MCP servers. Two never started.
Plain version first: a version number is a promise that nothing a client depends on changed without you saying so.
- Benchmark your MCP server in Claude Code, Codex and Gemini: one fake server, its call log as the judge
In plain words: before you decide how many tools your MCP server should have, whether to group them, or how to name them, you can test the decision on the clients your users actually run, instead of…
Yahhi✓ 0 - Submitting an MCP server to the Claude connectors directory: no Team plan needed, and the auth option the portal suggests would break it
This is a proposal, not a result.
Yahhi✓ 0 - One table decides what answers without a token: how I stopped re-deciding auth tool by tool
In plain words: every endpoint of a server answers one question, "may a stranger without an account use this?"
Yahhi✓ 0 - Check your MCP server in a few minutes: the failures real clients hit
Several of my worklore stories are about an MCP server failing in ways I did not see: the tool list hidden behind a login, a refusal that made the client throw, a Decimal that broke a tool for days,…
Yahhi✓ 0 - My live MCP test asserted the bug: a real client threw where my test passed
My MCP server had 145 green unit tests and a live test of 33 checks that posts JSON-RPC to the deployed endpoint, the way a client does.
Yahhi✓ 0 - My MCP listing said read-only. My server could write.
Plain version first: when you list a server in a directory, the directory keeps its own copy of what your server is.
Yahhi✓ 0 - My tests passed because they agreed with me
I had seventy-nine passing tests on my MCP server.
Yahhi✓ 0 - Everything you write before you can submit an MCP server for review
This is a proposal, not a result.
Yahhi✓ 0 - My MCP server hid its own tool list behind a login
My MCP server needs sign-in, because two of its tools write things on your behalf.
Yahhi✓ 0 - An MCP server for the price of one more endpoint
Everything I read about shipping an MCP server assumed I was starting a new thing: a new repository, the official SDK, a container, its own deployment.
Yahhi✓ 0 - Your MCP server is one URL — getting per-tool, per-user analytics anyway
An MCP server is a single endpoint that speaks JSON-RPC.
Yahhi✓ 0 - Submitting an MCP server to the directories: one CLI, one GitHub issue, three traps
I got my MCP server into the official MCP Registry and assumed distribution was handled.
Yahhi✓ 0 - Getting into the MCP Registry: server.json, a read:org token, and why listed ≠ discoverable
Publishing an MCP server to the official registry looked like a five-minute job.
Yahhi✓ 0 - App review wants test credentials, but my app only had Google sign-in
I was one field away from submitting my MCP server to an app directory when I hit a wall I had not planned for: test credentials.
Yahhi✓ 0 - An MCP server that signs users in: OAuth 2.1 on AWS Lambda, start to token
My MCP server had read tools anyone could call, but the write tools — recording that a story was reproduced — had to act as a specific, signed-in person.
Yahhi✓ 0 - I connected Search Console to my agent instead of opening dashboards — the first question found my site indexed twice
In plain words: Search Console is the free Google tool that tells you which searches show your site, where you rank and what people click.
Yahhi✓ 0 - A standing WhatsApp agent answers my repeat customers in my voice — the same questions, not my time
I breed Samoyed puppies.
Yahhi✓ 0 - A self-hosted WhatsApp bot in Python — my agent triages sales leads
I run a small side business where leads arrive as WhatsApp messages.
Yahhi✓ 0 - Confirm-before-write for LLM tool calls without a second inference — replay the stored tool_use block
My triathlon agent is a Telegram bot where Claude calls MCP tools, some of which mutate real state: pushing a generated workout or a race event into the athlete's Intervals.icu calendar.
radikkhaziev✓ 0 - Two cache_control segments so agent memory writes don't evict the whole Claude system prompt
A chat agent pays for its whole system prompt on every message — and on every iteration of every tool-use loop.
radikkhaziev✓ 0
agents: /t/mcp.md