sf-intelligence vs Salesforce MCP
Salesforce's 2026 MCP stack (Hosted MCP GA, DX MCP, Agentforce as client) made talking to your org from an AI client free and first-party. That is complementary to sf-intelligence — not a substitute for an offline dependency graph.
Short answer: Pitch resting only on "ask Salesforce from Claude/Cursor" is obsolete. Official servers cover live data and DevOps. Pitch sf-intelligence on graph depth, impact, permissions math, offline vault, and confidence labels — and use official MCP beside it when you need to act.
At a glance
| sf-intelligence | Salesforce MCP (hosted / DX) | |
|---|---|---|
| Maintainer | Community (MIT + Commons Clause) | Salesforce (official) |
| Primary purpose | Understand an org — impact, permissions, docs, exposure | Live data + DevOps — CRUD, SOQL, describe, deploy, retrieve |
| Write to org | No — read-only retrieve only | Yes on DX / hosted paths that expose mutate tools |
| Default mode | Offline local vault | Live API / Salesforce-hosted |
| Dependency graph | Yes — confidence-tagged edges | No dedicated cross-metadata graph |
| Impact / blast radius | Core feature | Not the documented focus |
| Permission tracing | Yes — FLS, sharing cascade | Platform permissions enforce API calls; not an offline cascade explainer |
| Agentforce | Companion audit + optional MCP server for agents | Native MCP client + registry / allowlists |
| Cost | Free | Included with eligible Salesforce editions / CLI |
| Package | sf-intelligence | Hosted setup + @salesforce/mcp / CLI |
Three official layers (and where we sit)
- Hosted MCP Servers — Salesforce-managed endpoints for Enterprise+ orgs: records, SOQL, describe-style metadata under platform auth.
- DX MCP Server — 60+ tools in opt-in toolsets for developers (metadata, deployment, data, LWC, code analysis, DevOps Center).
- Agentforce as MCP client — agents call allowlisted MCP tools; governance is the product story ("power without context bloat").
sf-intelligence is an ISV / community intelligence layer those clients can also call — or that humans use in Cursor/Claude Code before changing production.
Choose sf-intelligence when you need to…
- Ask "what breaks if I delete this field?" before anyone deploys
- Trace why a user cannot see a record or edit a field
- Document an inherited org or run a local agent-readiness audit
- Explore production metadata with zero write risk and no Salesforce round-trip per answer
- Cite provenance / confidence on every claim
Choose Salesforce MCP when you need to…
- Query or update live records under Salesforce auth
- Deploy, retrieve, run Apex/tests, scaffold LWC from an AI workflow
- Stay inside first-party hosting, Trust boundary, and admin allowlists
Use both together
Common pattern: sf-intelligence for understand, DX or Hosted MCP for implement. Register both in the same MCP client — they do not conflict. For Agentforce, treat sfi as a read-only intelligence server on the allowlist, not a replace-the-platform play.
More comparisons
See also: compare hub · Elements.cloud · dx0 · Hubbl · Metazoa · Sweep.
Disclaimer: Based on publicly documented Salesforce MCP capabilities as of mid-2026. Official offerings evolve — verify in Salesforce docs. sf-intelligence is not affiliated with Salesforce.
Add the graph layer.
Free on npm. Install beside Hosted MCP or DX MCP.