Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/gaffer-sh/mcp/agents-mdgit clone --depth 1 https://github.com/gaffer-sh/mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00615 | $0.00615 |
| Opus 5 | $0.00308 | $0.00308 |
| Sonnet 5 | $0.00123 | $0.00123 |
| Haiku 4.5 | $0.00061 | $0.00061 |
Grade A, and why
mcp AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md
MCP server for Gaffer. TypeScript, built with tsdown, published to npm as @gaffer-sh/mcp.
Architecture
This server is an external API consumer. It calls Gaffer's public /api/v1/ endpoints over HTTP like any third-party client, and never bypasses the public API. If an endpoint does not support what a function needs, the fix belongs in the API, not here.
The server runs in code mode: three MCP tools plus a codemode namespace of 16 functions that execute_code calls from sandboxed JavaScript. That keeps 16 tool definitions out of the client's context window and lets one execution chain several calls.
| Path | What it holds |
|---|---|
src/index.ts |
MCP server, the three tool registrations, stdio transport |
src/codemode/ |
Registry, executor, type generation, keyword search |
src/tools/ |
One file per function: inputSchema, execute, metadata |
src/api-client.ts |
HTTP client for the Gaffer API |
src/types.ts |
Shared types |
Commands
pnpm install
pnpm build # tsdown, then chmod +x dist/index.js
pnpm test # vitest
pnpm typecheck
Adding a function
- Create
src/tools/<name>.tsmatching the existing shape. - Add an API client method if needed.
- Register it in
src/codemode/register-tools.tswith a category and keywords. - Add it to the function table in
README.md.
Step 4 is enforced: src/codemode/__tests__/docs-sync.test.ts fails if a registered function is undocumented. The README is what npm and every MCP directory scrape, so it drifting is a distribution problem, not a docs nit.
Functions go in the codemode registry, not in the MCP tool list. Only add a first-class MCP tool when a client needs it before any code executes.
Environment
| Variable | Required | Description |
|---|---|---|
GAFFER_API_KEY |
Yes | gaf_ user key or gfr_ project token |
GAFFER_API_URL |
No | Defaults to https://app.gaffer.sh |
gaf_ keys read across projects and enable list_projects. gfr_ tokens are scoped to one project, and projectId resolves automatically, so callers must omit it.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 51 lines · 615 tokens per session scan A 0648f27e688f
mcp AGENTS.md is an instructions file published in the GitHub repository gaffer-sh/mcp (1 stars, last pushed 22d ago), licensed MIT. It adds 615 tokens to every session, about $0.0031 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
intervals-icu-mcp CLAUDE.md
Instructions for hhopke/intervals-icu-mcp, covering claude.md, project overview, development commands, architecture (quick reference) and tool categories.
ai-toolkit AGENTS.md
Instructions for pipefy/ai-toolkit, covering repository guidelines, documentation map, project structure, import namespace migration: pipefysdk → pipefy and src/pipefysdk/init.py (transitional shim).
flyto-core CLAUDE.md
Instructions for flytohub/flyto-core, covering claude notes, cross-agent handoff and shared code intelligence.
Plonk AGENTS.md
Instructions for ostapondo/Plonk, covering agent rules, layout, adding a module, build & verify and code style.
rosetta AGENTS.md
Instructions for tikoci/rosetta, covering codex instructions for rosetta, first reads, development defaults and mcp and client config.
pharo-smalltalk-interop-mcp-server CLAUDE.md
Instructions for mumez/pharo-smalltalk-interop-mcp-server, covering claude.md, project overview, development setup, environment variables and common commands.