The SaaS market bounced back. Now the interesting work happens at the edges.

Earlier this year, the iShares Expanded Tech-Software ETF (IGV) shed roughly $2 trillion in market cap as investors convinced themselves that AI would displace traditional B2B software wholesale. That recovery – now nearly complete at the index level – tells you something important: the market overreacted in both directions. AI isn’t replacing enterprise software. It’s being bolted onto it, wired into it, and increasingly, built around it through community-driven tools that don’t show up in any vendor’s press release.

The release of sf-mcp-server 0.1.0 on PyPI is a small but telling example of exactly that trend.

What sf-mcp-server Actually Does

It’s an independent, open-source MCP (Model Context Protocol) server for Salesforce – not a Salesforce product, and not affiliated with Salesforce in any way. It exposes Salesforce’s REST API, Bulk API 2.0, and Composite API to any MCP-compatible client, which in practice means you can connect AI agents and local tooling directly to your Salesforce org without going through a proprietary middleware layer.

That matters for a specific kind of team. If you’re running a RevOps function that needs custom data pipelines, bulk record operations, or tighter integration between AI workflows and your CRM, you’ve historically had three options: pay for an integration platform, hire a Salesforce developer to build something bespoke, or stitch it together with off-the-shelf connectors and hope they don’t break. sf-mcp-server is early – version 0.1.0 early – but it signals a fourth path that’s worth watching.

The practical use cases are concrete. Bulk-updating records based on signals from an external AI model. Pulling composite query results into a local agent that helps reps prioritise their sales pipeline. Running async data loads during off-hours without touching Salesforce’s UI. None of this is glamorous, but it’s the kind of operational plumbing that separates teams with clean data from those constantly fighting their own CRM.

Why the SaaS Recovery Doesn’t Tell the Whole Story

The index-level recovery is real. But it masks a significant divergence in how different types of software companies are actually performing in 2026.

Vendors that built tight AI integrations into their core product – rather than tacking on a chatbot – held up better during the selloff and recovered faster. The ones that didn’t are watching churn rate become a more pressing conversation. That’s the actual signal buried under the macro noise: customers are actively re-evaluating whether their existing CRM tools are keeping pace with what AI-native workflows now require.

For GTM teams, this creates a practical tension. You’re probably locked into Salesforce or HubSpot or whatever your company standardised on three years ago – you’re not ripping that out. But the teams outperforming their peers right now aren’t waiting for their CRM vendor to ship the perfect AI feature. They’re extending the platform themselves, through open-source tooling exactly like sf-mcp-server.

What This Means for Sales Teams Building Their Stack

There’s a broader pattern here worth taking seriously. The go-to-market stack has historically been a vendor-managed thing – you buy tools, configure them, and wait for the next release. That model is fraying. AI agents need programmatic access to CRM data that most vendor APIs weren’t designed to provide cleanly, and open-source projects fill that gap faster than product roadmaps do.

This doesn’t mean every RevOps team should be running their own MCP server. It means the teams with even one technically-capable person on the ops side have an advantage they didn’t have 18 months ago. The barrier to building custom CRM automation has dropped significantly, and the tooling ecosystem is growing around that fact.

For teams thinking about where to invest attention right now, a few things stand out:

If your sales cycle involves complex data transformations or high-volume record operations, programmatic CRM access is worth exploring – whether through a tool like sf-mcp-server or through your vendor’s native API.
AI agents that can read and write CRM data directly will change how reps interact with forecast tools – checking in on a sales forecast may eventually mean asking an agent rather than opening a dashboard.
Your win rate data is only useful if it’s clean and current. Bulk API tooling – the kind sf-mcp-server exposes – is exactly what keeps large orgs from drowning in stale records.

The open-source CRM tooling space is still early. Most of what’s being published now is at the 0.1.0 stage – functional, but not production-hardened. That’s fine. It’s worth tracking anyway, because the teams paying attention now will be the ones with working integrations when everyone else is still waiting on a vendor roadmap update.

The Takeaway for RevOps and CRM Professionals

The SaaS recovery story is mostly a distraction for practitioners. What actually matters is whether your stack can support the way AI is changing day-to-day sales work. Vendor platforms will eventually catch up – they always do. But the gap between what AI agents can theoretically do and what your CRM currently supports isn’t going to close itself.

If you’re serious about staying current, it’s worth bookmarking the CRM Daily Newsletter for ongoing coverage of exactly these kinds of developments – the unglamorous, infrastructure-level shifts that end up mattering more than the headline product launches.

The $2 trillion selloff that kicked off 2026 was a market overreaction. But the underlying anxiety driving it – that traditional SaaS wasn’t moving fast enough to keep pace with AI – wasn’t wrong. Open-source projects like sf-mcp-server are one answer to that anxiety: small, specific, and quietly useful. Which is how the tools that actually stick around get started.

Why Open-Source Salesforce Tools Are Having a Moment

CRM Daily Digest

Get this kind of intelligence every morning.