The whole surface is eight tools, which is a reasonable answer for anyone who wants an assistant to answer questions about their Salesforce data without adopting a large integration. Two things follow from the scope: update and delete are in the set, so the access token you supply is the limit on what can be changed, and the recommended order — describe the object before creating against it — is worth following, since Salesforce's reference fields are not guessable.
A lightweight Salesforce MCP server giving an assistant access to Salesforce objects and records. The author positions it as suited to prototyping and small projects rather than as a complete platform integration.
- Query tools — `soql_query` to execute SOQL, `search_records` to search across multiple objects with a limit and pagination, and `get_record` to retrieve one record by ID
- CRUD on records: `create_record`, `update_record` and `delete_record`
- Metadata to build queries against — `describe_object_definition` for object metadata and field information with pagination, and `list_avail_objects` for the objects available in the org
- Guidance built into the tool descriptions: describe the object first so the reference fields are known before creating a record, and confirm before UPDATE and DELETE operations
- Built-in rate limiting alongside the Salesforce authentication
A Salesforce org — developer or production — with a Connected App configured, from which you take the two required environment variables: `SALESFORCE_ACCESS_TOKEN` and `SALESFORCE_INSTANCE_URL`. Python 3.10 or higher. The published package is `mcp-salesforce-lite` on PyPI; the recommended client configuration runs it with `uvx --from mcp-salesforce-lite mcp-salesforce-lite`. For development, `uv sync` or `poetry install` from a clone. MIT licensed.
One command plus a key — uvx --from mcp-salesforce-lite mcp-salesforce-lite, then supply credentials