simple-salesforce-mcp
Enables users to connect to Salesforce and perform CRUD operations, run SOQL/SOSL queries, and manage records through a lightweight MCP server with built-in safety gates for writes.
README
simple-salesforce-mcp
A lightweight Salesforce MCP server. Talks straight to the
Salesforce REST API over stdio with two runtime dependencies (mcp, httpx) — built as a
fast-cold-start replacement for the official @salesforce/mcp DX server, whose dependency
tree is too heavy for sandboxed environments.
Tools
| Tool | Writes? | Notes |
|---|---|---|
run_soql_query |
no | Raw SOQL escape hatch with pagination (next_url) |
search_records |
no | Cross-object full-text search (parameterized search API) |
get_record |
no | Fetch one record by Id, optional field list |
create_record |
yes | Create a record |
update_record |
yes, confirm-gated | Requires literal confirm=true |
delete_record |
yes, confirm-gated | Requires literal confirm=true |
describe_object |
no | Trimmed object metadata (fields, types, required, picklists) |
list_objects |
no | Org objects, system noise filtered by default |
get_org_info |
no | Current user, org, instance URL, API version |
Every tool carries MCP annotations (readOnlyHint / destructiveHint / idempotentHint).
update_record and delete_record refuse to act unless the call includes confirm: true,
and the refusal message instructs the model to present the exact change to the user and get
approval first. SOQL itself cannot modify data, so the escape hatch stays read-only.
Authentication
The server performs no OAuth flow itself; it consumes an existing access token, resolved on every tool call (so an external process may rotate the token at any time):
-
Salesforce CLI auth store under
$HOME:~/.sf/config.json→{"target-org": "<username>"}(or legacy~/.sfdx/sfdx-config.json→{"defaultusername": ...})~/.sfdx/<username>.json→{"accessToken": "...", "instanceUrl": "https://..."}
This is the shape written by
sf org login, and the shape the JustParent platform materialises into sandboxes (with the platform refreshing the access token before each call). Refresh tokens and client secrets in these files are ignored — never read into memory, stored, or logged. -
Environment variables (fallback, for local/standalone use):
SALESFORCE_ACCESS_TOKENandSALESFORCE_INSTANCE_URL.
On a rejected token (HTTP 401 / INVALID_SESSION_ID) the server re-reads the auth store
once and retries, then reports that the connection needs to be re-established.
The REST API version defaults to 62.0; override with SALESFORCE_API_VERSION.
Running
# Standalone with env vars (get a token via: sf org display --json)
SALESFORCE_ACCESS_TOKEN=... \
SALESFORCE_INSTANCE_URL=https://yourorg.my.salesforce.com \
uvx --from git+https://github.com/JustParent/simple-salesforce-mcp simple-salesforce-mcp
Claude Desktop / generic MCP client config:
{
"mcpServers": {
"salesforce": {
"command": "uvx",
"args": ["--from", "git+https://github.com/JustParent/simple-salesforce-mcp", "simple-salesforce-mcp"],
"env": {
"SALESFORCE_ACCESS_TOKEN": "...",
"SALESFORCE_INSTANCE_URL": "https://yourorg.my.salesforce.com"
}
}
}
}
Smoke test (no network, prints version, credential status, and tools; exits 0):
uvx --from . simple-salesforce-mcp --test
Development
uv run pytest # unit tests, no network needed
uv run ruff check . # lint
uv run ruff format . # format (--check to verify only)
uv run mypy # type check (src and tests)
uv run simple-salesforce-mcp --test
npx @modelcontextprotocol/inspector uv run simple-salesforce-mcp # interactive
CI (.github/workflows/tests.yml) runs on every push and pull request: the test
suite on Python 3.10–3.13, ruff lint and format checks plus mypy, and a packaging
job that installs the package the way sandboxes do (uvx --from .) to catch
packaging breakage.
Security notes
- Access-token only: the server never requests, reads, or persists refresh tokens.
- Destructive operations (update/delete) are gated behind an explicit
confirm=trueargument with model-facing guidance to obtain user approval first. - SOQL pagination cursors (
next_url) are validated to be/services/data/...paths, and object/record identifiers are validated before being placed in URLs. - Responses are size-capped with explicit truncation notices.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。