Introduction
This is the third post in a series. The first ranked developer experience across 41 technology marketplaces. The second mapped MCP server maturity across those same 41. One finding from the second post kept nagging at us: Docs MCP servers (the read-only servers that make a platform's documentation queryable by AI coding assistants) were the single most underinvested category. Only 8 out of 41 major marketplaces had one. We called it the lowest-hanging fruit in the MCP landscape and moved on.
Then we went looking outside the 41.
We wanted to understand who IS shipping Docs MCP servers beyond the big marketplaces, what they have in common, and whether the gap was as bad as it looked. What we found surprised us. The biggest wave of Docs MCP adoption isn't happening company-by-company. It's happening platform-by-platform, driven by four documentation providers that auto-generate MCP servers for all of their customers simultaneously. The result: roughly 19,000 companies now have a Docs MCP server. Most of them didn't build it, didn't ask for it, and probably don't know it exists.
The Documentation Platform Multiplier
Four documentation platforms now auto-generate Docs MCP servers for every customer on their platform.
| Platform | Customers Enabled | MCP Endpoint Pattern | Details |
|---|---|---|---|
| Mintlify | ~10,000 companies | /mcp | Auto-generates for all customers. Covers API references and tutorials. |
| GitBook | ~6,500 companies | /~gitbook/mcp | Every published GitBook site gets MCP automatically. Announced September 2025. |
| ReadMe | ~2,500 companies | /mcp | Out-of-box MCP server for every project. Launched during Launch Week 2025. |
| Fern | ~150 companies | /_mcp/server | Auto-generates from OpenAPI specs. Customers include Square, ElevenLabs, Cohere, Webflow, LaunchDarkly. |
Total: approximately 19,000 companies with Docs MCP servers, the vast majority of which were enabled without any action from the company itself.
This is the documentation platform multiplier effect, and it has a direct precedent. In November 2024, Mintlify enabled llms.txt support (the protocol for making documentation AI-readable) and thousands of companies including Anthropic, Cursor, Pinecone, and Windsurf got AI-readable docs overnight. No configuration required. No engineering ticket filed. The docs platform flipped a switch and the entire customer base moved forward.
The same playbook is now running with MCP, at a larger scale and with deeper integration. llms.txt gave AI models a static text file to read. MCP gives AI agents a queryable server that can search, filter, and retrieve documentation contextually. It's the difference between handing someone a printed manual and giving them a librarian who knows the collection.
This is also the answer to a question we asked in our MCP maturity post: why don't more companies have Docs MCP servers? Many do. They just didn't build them. Their docs platform did it for them.
One additional note: Fern was acquired by Postman in late 2025. That puts MCP auto-generation capability inside one of the largest API tooling platforms in the world. If Postman rolls Fern's MCP generation across its broader customer base, the 19,000 number could move significantly. Worth watching.
The 35 Companies Building Their Own
Outside the documentation platform multiplier, we found 35 companies that have built custom Docs MCP servers, API MCP servers, or both. These are companies not among the 41 major marketplaces we ranked in our previous posts, and not relying on Mintlify, GitBook, ReadMe, or Fern for auto-generation. They chose to invest engineering time in MCP deliberately.
The vertical breakdown tells a clear story.
| Vertical | Companies | Count |
|---|---|---|
| Databases | Supabase, Neon, PlanetScale, Turso, Prisma, Redis, MongoDB, Pinecone, Weaviate | 9 |
| AI/ML Platforms | Hugging Face, LangChain, LlamaIndex, Mistral AI, Replicate, Cohere | 6 |
| CMS Platforms | Sanity, Contentful, Webflow, Payload CMS, Contentstack | 5 |
| Dev Tools & Workflow | Linear, LaunchDarkly, n8n, Retool, Convex, Appwrite | 6 |
| Auth & Identity | Clerk, Auth0, Stytch | 3 |
| Infrastructure & Hosting | Cloudflare, Railway, Render | 3 |
| Analytics & Monitoring | PostHog, Sentry, Mixpanel | 3 |
Here are the patterns worth noting.
Databases: The Most Aggressive Vertical
Nine out of 35 companies are databases. That's disproportionate by any measure. Supabase, Neon, PlanetScale, Turso, Prisma, Redis, MongoDB, Pinecone, and Weaviate have all shipped MCP servers, most with both docs and API coverage.
The hypothesis is straightforward: database companies understand that AI coding assistants need to write correct queries and schema definitions, and stale training data causes the most damage in their domain. A wrong API call to a CRM fails gracefully. You get an error response and try again. A wrong database migration doesn't fail gracefully. It corrupts data, drops columns, or creates schema drift that takes days to untangle. When your product is the system of record, you have an outsized incentive to make sure the AI assistant has current information.
The implementation approaches vary. Supabase and Neon both ship combined docs-plus-API MCP servers. PlanetScale built a custom server with OAuth, announced January 2026. Turso took a different approach entirely, building MCP directly into their CLI with a --mcp flag. Weaviate uses Kapa.ai specifically for their docs MCP, while maintaining a separate custom API MCP server. Pinecone went furthest, shipping three separate MCP servers: an Assistant remote server, an Assistant local server, and a Developer local server.
The database vertical has moved from early adoption to table stakes. If you're a database company without an MCP server in 2026, you're behind your peer set.
AI/ML: Building for Their Own Users
Hugging Face, LangChain, LlamaIndex, Mistral, Replicate, and Cohere have all shipped MCP servers. This makes intuitive sense: these companies are also building the tools that consume MCP servers. Their developer communities are the earliest and most sophisticated MCP users. They have the deepest understanding of the protocol and the strongest incentive to make their own docs AI-queryable.
Hugging Face's implementation is notable: a hosted MCP server at huggingface.co/mcp that spans their entire model hub documentation. LangChain includes an mcpdoc tool specifically for querying documentation through llms.txt, bridging the static and dynamic approaches. Mistral integrated MCP directly into their Python SDK, making it a first-class part of their agent framework rather than a separate service.
CMS: The Content-to-MCP Pipeline
Five content management platforms (Sanity, Contentful, Webflow, Payload CMS, and Contentstack) have shipped MCP servers. Content management is fundamentally about structured data, and MCP is a natural fit for querying it programmatically.
Sanity went deep, shipping a hosted MCP server at mcp.sanity.io that's GA in 2026. Webflow took the build-versus-buy decision literally and chose buy: they use Fern for auto-generation rather than building custom. That's a practical choice and one that other companies should evaluate. If your documentation is already structured for a generator, the engineering cost of a custom MCP server may not be justified.
Contentstack's implementation is the outlier: experimental and not officially supported. It's a reminder that "has an MCP server" and "has a production-ready MCP server" are different statements. If you're depending on a platform's MCP server for your workflow, check whether it's GA or experimental before building against it.
Auth: Where Stale Docs Create Security Vulnerabilities
Clerk, Auth0, and Stytch have all shipped MCP servers. Auth is one of the domains where a developer following stale guidance creates a security vulnerability, not a bug. Deprecated OAuth flows, outdated token handling patterns, incorrect scope configurations ... these are attack surfaces.
A Docs MCP server for an auth platform is a safety mechanism as much as a convenience. When a developer's AI assistant generates auth code using current documentation instead of 18-month-old training data, the code is more likely to be correct and secure. Auth0's implementation uses OAuth Device Authorization Flow for the MCP server itself, which is a nice bit of dogfooding.
Infrastructure, Analytics, and Dev Tools
The remaining companies span infrastructure (Cloudflare, Railway, Render), analytics (PostHog, Sentry, Mixpanel), and developer tools (Linear, LaunchDarkly, n8n, Retool, Convex, Appwrite).
A few standouts. Cloudflare built their MCP server on Workers, using their own platform as the hosting layer. PostHog ships 27 tools across 7 categories in their MCP server, hosted at mcp.posthog.com, one of the most comprehensive implementations we found. Sentry launched both an MCP server for error monitoring AND an MCP Server Monitoring tool, meaning they're helping other companies monitor the health of their MCP servers. Linear's remote server at mcp.linear.app/mcp was one of the earlier implementations, announced May 2025.
Appwrite deserves a specific mention for shipping two separate MCP servers: one for API operations and one for docs. That's the architecture we recommended in our framework, and they're one of the few companies outside the big 41 that implemented it as distinct, purpose-built servers.
The Adopter Profile
Synthesizing across all 35 companies, a clear profile emerges.
By stage. Roughly 40% are Series A-B startups (Neon, Turso, Clerk, Convex, Railway, Render). About 30% are Series C-D growth companies (Supabase, PlanetScale, Sanity, Contentful, LaunchDarkly). The remaining 25-30% are public or enterprise companies (Redis, MongoDB, Cloudflare, Sentry, Elastic). Startups are adopting first because MCP is a competitive differentiator when you're fighting for developer attention. A Series B database company can't outspend MongoDB on marketing, but it can out-invest on developer experience. A well-built MCP server is a high-leverage move for a small team.
By audience. Nearly every company on this list sells to developers. The exceptions (Plaid and Modern Treasury in fintech) still serve a primarily developer audience through their APIs. Companies selling to non-developers haven't moved on MCP in meaningful numbers. This makes sense: the MCP audience today is developers using AI coding assistants. When non-technical users start interacting with MCP through products like ChatGPT or Claude, the adopter profile will broaden. But today, this is a developer-tools story.
By approach. Most of the 35 ship API MCP and Docs MCP together or as part of a unified server. The "API only, skip docs" pattern we observed in the big 41 marketplaces is not the pattern here. Smaller companies that build their own MCP tend to do both. This is a telling signal. The "skip docs" approach in the big marketplaces isn't a deliberate architectural choice. It's a prioritization failure driven by larger organizations where the docs team and the API team don't coordinate on MCP as a shared initiative. Smaller, more focused teams naturally build both because the same person or team owns the entire developer experience.
By timing. Virtually every implementation on this list was announced in the last 12-15 months. MCP was announced by Anthropic in November 2024. By mid-2025, early adopters were shipping servers. By early 2026, the vertical we mapped above was in place. The protocol went from announcement to widespread adoption in under 18 months. For context, GraphQL took roughly 3-4 years to reach comparable adoption in developer tooling. MCP is moving faster because it's riding the AI coding assistant wave. The demand side was already there waiting.
What This Means
If you use Mintlify, GitBook, ReadMe, or Fern for your documentation, you probably already have a Docs MCP server. Check your docs URL with the endpoint patterns listed above. If it responds, your documentation is already queryable by AI coding assistants. Promote it to your developer community. Add it to your docs landing page. Mention it in your changelog. The server exists; the distribution is up to you.
If you're a database company without MCP, you are now behind 9 of your peers. This vertical has moved past early adoption. Supabase, Neon, PlanetScale, Turso, Prisma, Redis, MongoDB, Pinecone, and Weaviate have all shipped. The window where MCP was a differentiator for database companies is closing. It's becoming an expectation.
If you're evaluating which platforms to build integrations on, check their MCP status. A platform with an official MCP server means your integration can be discovered and consumed by AI agents. A platform without one means you're limited to traditional distribution channels: human users navigating a marketplace UI. As AI agents become a primary interface for developer workflows, the platforms with MCP servers will capture disproportionate integration traffic. The gap is widening quarterly.
The build-versus-buy decision for Docs MCP is now trivially answered. If your docs are on a platform that auto-generates MCP, it's already done. If your docs are custom-built, tools like Twilio's open-sourced OpenAPI-to-MCP generator and Fern's auto-generation framework mean building a Docs MCP server is a matter of hours or days, not weeks. The engineering cost is no longer a valid reason to delay.
Conclusion
The Docs MCP gap we identified across the 41 major marketplaces is real, but it's closing ... just not the way most people expected. It's not individual companies making deliberate investment decisions one by one. It's documentation platforms enabling MCP at infrastructure scale, turning on the capability for thousands of companies simultaneously. The 19,000 number reflects a market that's moving top-down through platform providers, not bottom-up through individual engineering teams.
The 35 companies building their own tell the other half of the story. They're overwhelmingly developer-tool companies, disproportionately databases, and they skew toward startups that treat developer experience as a competitive weapon rather than a cost center. They ship Docs MCP and API MCP together because they understand something the larger platforms have been slower to grasp: when your customer's AI assistant has stale information about your product, every wrong suggestion is a support ticket you caused. Every hallucinated API parameter is a frustrated developer who blames your platform, not the AI.
Taken together with the DX ranking and the MCP maturity analysis, the picture is consistent. The platforms that invest in developers win on DX rankings, lead on MCP adoption, and now either ship their own Docs MCP or have their docs platform do it for them. The platforms that don't invest in developers fall behind on all three dimensions simultaneously. The compounding effect is real, and 2026 is the year it becomes visible in market share.
At Left Hook, we help companies on both sides of this: building integrations against platforms with MCP servers, and helping product companies ship their own. If you're looking at this data and wondering where your company fits, or if your docs platform already gave you an MCP server you haven't promoted yet, we can help you figure out the next move.
This analysis reflects Docs MCP server availability as of February 2026. Company counts for documentation platforms are estimates based on publicly available data and may vary. We will update this post as the landscape evolves.