apitameBLOCK

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

npx apitame init ./openapi.yaml          # or an API URL, postgres://..., mysql://..., ./shop.db
npx apitame add-to claude-code --write   # or codex, cursor, vscode, gemini, claude-desktop

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: github-api
description: Use the GitHub REST API (subset) through the "github" MCP server (generated by apitame). Use for requests that involve repos, issues, pulls, search, users or activity data (read-only access). Covers which tool to call, argument shapes, common multi-step workflows and the safety rules for writes.
---

# GitHub REST API (subset)

A hand-picked subset of the GitHub REST API (repositories, issues, pull requests, search, users and stars), used by the apitame examples. Operation ids, paths and fields follow the official GitHub description.

These tools come from the `github` MCP server, generated by apitame from the OpenAPI spec.
The server runs in **read-only mode**: 13 write operations are blocked.

## Tools

| Tool | Use it to | Operations |
| --- | --- | --- |
| `repos` | Repositories, their content, commits, branches and releases | `repos/get`, `repos/list-for-user`, `repos/list-for-org`, `repos/list-contributors`, `repos/list-languages`, `repos/list-branches`, `repos/list-commits`, `repos/list-releases`, `repos/get-content`, `repos/get-commit`, `repos/get-latest-release` |
| `issues` | Issues, comments and labels | `issues/list-for-repo`, `issues/list-labels-for-repo`, `issues/get`, `issues/list-comments` |
| `pulls` | Pull requests, their files and reviews | `pulls/list`, `pulls/get`, `pulls/list-files`, `pulls/list-reviews` |
| `search` | Search repositories, issues and code | `search/repos`, `search/issues-and-pull-requests`, `search/code` |
| `users` | User profiles | `users/get-authenticated`, `users/get-by-username` |
| `activity` | Stars and stargazers | `activity/list-stargazers-for-repo` |
| `search_operations` | Find any of the 38 operations by keyword | all |
| `call_operation` | Call an operation that has no dedicated tool | any available |

Grouped tools take `{"operation": "<operation id>", "params": {...}}`. Results are JSON: `{"status": <http status>, "data": <body>}`.

Add `"fields"` with dot paths to keep only what you need, for example `"fields": ["id", "name", "items.owner.login"]`. Arrays are handled for you, so `items.name` keeps `name` in every item. It saves a lot of context on large responses.

## Workflows

### List commits, then open one

1. Call `repos` with `{"operation":"repos/list-commits","params":{"owner":"octocat","repo":"hello-world"}}`.
2. Take the identifier (`ref`) of the item the user means from the result, then call `repos` with `{"operation":"repos/get-commit","params":{"owner":"octocat","repo":"hello-world","ref":"main"}}`.
3. Summarize the fields the user asked about instead of pasting the raw JSON.

### List repository issues, then open one

1. Call `issues` with `{"operation":"issues/list-for-repo","params":{"owner":"octocat","repo":"hello-world","state":"open"}}`.
2. Take the identifier (`issue_number`) of the item the user means from the result, then call `issues` with `{"operation":"issues/get","params":{"owner":"octocat","repo":"hello-world","issue_number":1}}`.
3. Summarize the fields the user asked about instead of pasting the raw JSON.

### Search repositories

1. Call `search` with `{"operation":"search/repos","params":{"q":"example"}}`. Find repositories via keywords and qualifiers, for example "mcp language:typescript stars:>100".
2. Useful parameters: `q`, `sort`. Ask the user for missing values instead of guessing.

### Page through long lists

1. List operations such as `repos/list-for-user` accept `per_page`, `page`. Start small, and only fetch the next page if the user needs more.
2. Large responses are cut to fit the response limit; a "note" field says when that happened.

## Safety rules

- Write operations are blocked by the server. Do not look for workarounds. If the user needs one, name the operation and tell them it must be enabled in `apitame.config.yaml` (`mode: read-write` and `writes.allow`).
- Responses over 50 KB are cut. Use filters, pagination and `fields` rather than fetching everything.
- Never ask the user for API keys or tokens: the server already holds them, and they are never shown to you.
- A non-2xx `status` is an API error: read `data` for the reason before retrying.

## Reference

Every operation with its parameters: [references/operations.md](references/operations.md).

BLOCKgrade F · trust 56/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

What the source does

Filesystem

declared (5 observation(s))

Network

declared (3 observation(s))

Shell

declared (2 observation(s))

Dependencies

not all pinned

Secrets in source

found

Findings (25)

# DATABASE_URL=postgres://apitame_ro:password@localhost:5432/shop

url: `http://127.0.0.1:${port}${basePath}`,

APITAME_TEST_POSTGRES_URL: postgres://postgres:[email protected]:5432/postgres

APITAME_TEST_MYSQL_URL: mysql://root:[email protected]:3306/mysql

npx apitame init postgres://readonly_user:password@localhost:5432/shop

npx apitame init mysql://readonly_user:password@localhost:3306/shop

const SECRET = 'live-secret-token-0123456789';

const SECRET = 'fake-key-ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789';

echo: { token: 'unregistered-9f8e7d6c' },

.prettierignore

.prettierrc.json

import { loadConfig } from '../../config/load.js';

import { zip } from '../../generator/zip.js';

import { loadSkills } from '../../server/skills.js';

import { VERSION } from '../../version.js';

import { envRefs, loadDotEnv } from '../../config/env.js';

Prefer HTTP? `npx apitame serve --transport http` listens on `http://127.0.0.1:3333/mcp` (add `--sessions` so destructive actions are approved in the client). More in [docs/clients.md](docs/clients.md

apitame serve --transport http --port 3333     # streamable HTTP on http://127.0.0.1:3333/mcp

# MCP endpoint: http://127.0.0.1:3333/mcp, health check: http://127.0.0.1:3333/health

Claude Code: `claude mcp add --transport http github http://127.0.0.1:3333/mcp`.

@modelcontextprotocol/node, @modelcontextprotocol/server, @readme/openapi-parser, ajv, commander, mysql2, pg, yaml

- **Prompt injection in data.** apitame does not filter what an API or table returns. If a record says "ignore previous instructions", the agent sees it. The safety layers make sure such instructions

| [PrefectHQ/fastmcp](https://github.com/PrefectHQ/fastmcp) (`FastMCP.from_openapi`) | 28,067 | 2026-10-13 | Python | Runtime proxy inside a Python framework. Every route becomes a tool by default; `R

| [harsha-iiiv/openapi-mcp-generator](https://github.com/harsha-iiiv/openapi-mcp-generator) | 639 | 2026-09-26 | TypeScript | Codegen: scaffolds a Node project with one tool per operation and zod vali

| [Stainless](https://www.stainless.com/docs/mcp/index.md) (closed source) | n/a | n/a | TypeScript (generated) | Codegen alongside SDKs. Current default is code mode: a code execution tool plus a doc

Gates applied: no_behavioural_pass.