# TechFabric's MCP servers > TechFabric runs two MCP servers. www.techfabric.com/mcp searches and reads the site anonymously and takes blog drafts from TechFabric authors, and databricks.techfabric.com/mcp serves the Databricks Field Guide, public and read-only. This page has every tool, resource and limit, and the setup for Claude Code, Claude Desktop, Cursor and VS Code. Canonical: https://www.techfabric.com/mcp-servers --- MCP servers ## TechFabric's MCP servers Coding agents write Databricks code from training data that's months old. TechFabric's engineers build and run Databricks platforms for clients and write down what works in the Databricks Field Guide. Its MCP server lets Claude Code, Cursor or VS Code read the right section before writing code. The agent gets the explanation, the code that runs, and the docs.databricks.com pages to check against. - Free - No sign-in to read - One line to connect ### Which MCP servers does TechFabric run? Two free servers for engineers using Claude Code, Claude Desktop, Cursor or VS Code: the Databricks Field Guide at databricks.techfabric.com/mcp, for Databricks answers with working code, and techfabric.com at www.techfabric.com/mcp, for our blog, services and case studies. Add either URL to your client as a Streamable HTTP server; reading needs no sign-in. #### Databricks Field Guide MCP server 5 tools Search and read TechFabric's Databricks Field Guide: chapters, code examples and what changed. Chapter and code answers link to the docs.databricks.com pages their sections cite. - Public, read-only - No sign-in - Protocol 2026-07-28 - Free MCP Registry entry Glama listing #### techfabric.com MCP server 8 tools Search and read TechFabric's Databricks and AI content: blog posts with FAQs, services, case studies. TechFabric authors can also draft blog posts from their own agent. - Read without sign-in - Sign in to write - Protocol 2026-07-28 - Free MCP Registry entry Glama listing The Field Guide server keeps nothing your agent asks, and connecting either server is one line of configuration. How it works ### How does an AI agent use an MCP server? The Model Context Protocol (MCP) is the open standard AI assistants use to call tools on other systems. An agent connected to an MCP server can look something up halfway through a task and get a structured answer back, with a URL to cite, where it would otherwise guess from its training data or scrape a web page. Diagram: One agent, two servers. Reading either one needs no sign-in; only the techfabric.com write tools ask a TechFabric author to sign in with Microsoft, and nothing on either server publishes. Databricks Field Guide ### What tools does the Databricks Field Guide MCP server have? Five read-only tools. They search the guide, read a chapter or one section, find code examples, list what Databricks changed lately and list the chapters. The Databricks Field Guide is what our engineers know about building, running and paying for Databricks. Ask how to add a Unity Catalog row filter, and the agent gets the section that explains it, the SQL function from that section, and the docs.databricks.com pages the section cites. Databricks' documentation stays the authority on syntax and limits. **Sign-in** None. The server is public and read-only, and it has no write tools. **Protocol** MCP revision 2026-07-28 over Streamable HTTP. Clients still on 2025-11-25 are served on the same URL. **Rate limits** 60 calls a minute and 300 an hour per IP address. **Freshness** Rebuilt on every deploy of the guide. Every answer carries the commit that built it, each page's last-changed date and how long it may be cached, which is 10 minutes for what's new, 15 for search and an hour for chapters and code. **Registry** com.techfabric.databricks/field-guide, version 1.1.1. MCP Registry entry · Glama listing #### The Field Guide's five tools search_guide Read ##### Search the guide The guide sections that answer a Databricks question. **Returns and parameters** **Returns** The chapter sections, margin notes and events that answer a Databricks question, best first, each with a snippet, its URL with the section anchor, and when the page last changed. **Parameters** query (required); kind: chapter, note, feature or event; limit 1 to 25, default 10 get_chapter Read ##### Read a chapter A chapter or one section as Markdown, code intact. **Returns and parameters** **Returns** A chapter or margin note as clean Markdown with its code intact, or one section of it, plus the outline, the code blocks listed separately, the FAQs, the dates and the docs.databricks.com links it cites. **Parameters** slug (required): a slug such as lakebase, a path or any guide URL; section: an anchor or a heading get_code_examples Read ##### Find code examples SQL, Python, CLI and config blocks for a topic. **Returns and parameters** **Returns** SQL, Python, CLI and config blocks that match a topic, each with the explanation around it, the section it came from and the docs links that section cites. **Parameters** topic (required); lang such as sql, python, bash, yaml, json or hcl; limit 1 to 25, default 8 whats_new Read ##### What's new in Databricks Recent Databricks changes the guide tracks, newest first. **Returns and parameters** **Returns** Recent Databricks changes the guide tracks, newest first: margin notes, release-stage changes with links to Databricks' documentation and release notes, and chapters updated. **Parameters** since (YYYY-MM-DD, default the last 30 days); limit 1 to 200, default 30 list_chapters Read ##### List chapters The table of contents, with each chapter's main sections. **Returns and parameters** **Returns** Every chapter in the guide's order, with its part, summary, dates, URL, Markdown URL and main sections, and the margin notes. **Parameters** None #### Field Guide resources fieldguide://llms.txt Every chapter, note and feature page, one line each. fieldguide://page/{+path} Any published page as Markdown, for example fieldguide://page/guide/lakebase. techfabric.com ### What tools does the techfabric.com MCP server have? Eight. Anyone can call the four read tools, which search the site, read a page, list the blog and list our services. The other four are for TechFabric's blog authors, who check and file drafts with them. Use it when an agent needs something TechFabric has published, such as a blog post on Lakebase, the FAQs on a service page, or the list of what we offer. It answers only from published, indexable pages, so drafts, previews and anything marked noindex never appear in a result. **Sign-in** None to read. The write tools are for TechFabric blog authors, who sign in with their Microsoft (Office 365) account in the techfabric.com tenant. No other tenant is accepted. **Protocol** MCP revision 2026-07-28 over Streamable HTTP. Clients still on 2025-11-25 are served on the same URL. **Rate limits** 300 anonymous calls per IP address per hour. Signed-in calls are counted per author, at 600 an hour, with at most 20 new check runs, 6 submissions and 30 saves. **Freshness** The index is rebuilt from the site on every deploy. Resources may be cached for an hour, and search results for 15 minutes. **Registry** com.techfabric/techfabric. MCP Registry entry · Glama listing #### techfabric.com's eight tools Author tools work only for TechFabric authors signed in from Claude Code or a Claude custom connector, and there's no publish tool. The workflow is under writing a TechFabric blog post with your own agent . search Read ##### Search techfabric.com Matching pages, best first. **Returns and parameters** **Returns** Blog posts, service and solution pages, case studies, city pages and events that match, best first, each with its title, URL, summary, date, author and type. **Parameters** query (required); type: post, service, solution, case_study, city, event or page; limit 1 to 25, default 10 get_page Read ##### Read a page One published page as clean Markdown, with its FAQs. **Returns and parameters** **Returns** One published page as clean Markdown, with its title, search title, description, dates, author, tags, its FAQs as question and answer pairs, and its canonical URL. **Parameters** url_or_slug (required): a full URL, a path such as /services/databricks, or a post's slug list_posts Read ##### List blog posts Blog posts, newest first, by tag, author or date. **Returns and parameters** **Returns** Blog posts, newest first. **Parameters** tag, author (a name or a slug), since (YYYY-MM-DD); limit 1 to 100, default 20 get_services Read ##### What TechFabric offers Every service and product page, with its own summary. **Returns and parameters** **Returns** Every service, solution, Databricks, AI, cloud, software, accelerator and product page, with its name, URL, kind and summary exactly as the live page states them. **Parameters** None check_draft Author ##### Check a draft Findings on a draft, each with the passage it's about. **Returns and parameters** **Returns** Runs the site's own checks on a post and returns each item's status and findings, with the passage each finding is about. Files nothing. A full run takes five to ten minutes. **Parameters** markdown (required, with frontmatter); images; wait_seconds up to 50, default 25 submit_draft Author ##### Submit a draft A draft filed under your byline, with its preview. **Returns and parameters** **Returns** Files the post under the signed-in author's byline exactly as written, builds a preview and emails the author the editor link. Returns the draft's id. **Parameters** title and markdown (required); images get_status Author ##### Where a draft stands A draft's state, failing checks, base_sha and links. **Returns and parameters** **Returns** One of your own drafts: its state, what's failing, the last check run, its current Markdown, and the preview and editor links. **Parameters** id (required) update_draft Author ##### Save an edit to your draft A new saved version of your draft, preview rebuilt. **Returns and parameters** **Returns** Saves a new version of your own draft and rebuilds the preview. It runs no checks; those run from the editor. **Parameters** id, markdown and base_sha (all required; base_sha comes from get_status) #### techfabric.com resources site://llms.txt The site's map for language models. site://page/{+path} Any published page as Markdown, for example site://page/services/databricks. site://search?query= The search tool's results as a resource a client can cache, with an optional type. techfabric://editorial/rules The house writing rules an author's agent reads before it drafts. Connect ### How to connect Claude, Cursor or VS Code to TechFabric's MCP servers Add the server's URL to your client. Both servers are remote, so there's nothing to install, and the names below are only labels. #### How to connect Claude Code Run each command once, in any terminal. Add --scope user to have it in every project. ``` claude mcp add --transport http techfabric https://www.techfabric.com/mcp ``` ``` claude mcp add --transport http databricks-field-guide https://databricks.techfabric.com/mcp ``` #### How to connect Claude Desktop and claude.ai - Open Settings, then Connectors, and choose Add custom connector. - Give it a name and paste the server's URL. Leave the OAuth fields empty. - Choose Add. Repeat for the second server. On a Team or Enterprise plan, an owner may have to add a custom connector for the organisation first. #### How to connect Cursor Add this to ~/.cursor/mcp.json for every project, or to .cursor/mcp.json in one project. ``` { "mcpServers": { "techfabric": { "url": "https://www.techfabric.com/mcp" }, "databricks-field-guide": { "url": "https://databricks.techfabric.com/mcp" } } } ``` #### How to connect VS Code Add this to .vscode/mcp.json in the workspace, then start the servers from that file or from the MCP view. The tools are there in Copilot's agent mode. ``` { "servers": { "techfabric": { "type": "http", "url": "https://www.techfabric.com/mcp" }, "databricks-field-guide": { "type": "http", "url": "https://databricks.techfabric.com/mcp" } } } ``` #### How to connect other MCP clients Point the client at the URL with the Streamable HTTP transport and no authentication. A client that only speaks stdio can reach either server through the mcp-remote bridge, with this as its command. ``` npx mcp-remote https://databricks.techfabric.com/mcp ``` To try one by hand, a JSON-RPC tools/list POSTed to the URL returns the tools with their full input and output schemas. Reading never asks you to sign in. Writing to techfabric.com is for TechFabric's authors, and its sign-in and workflow are under writing a TechFabric blog post with your own agent . Try it ### What can you ask TechFabric's MCP servers? Anything the guide or the site covers. Name the server the first time, so the agent knows which tools to reach for; after that it usually picks them on its own. #### Prompts for the techfabric.com server " Search techfabric.com for what TechFabric has written about Lakebase and summarise the three newest posts, with links. " " Using the techfabric server, list TechFabric's Databricks services and say in a line what each one covers. " " Read techfabric.com/durable-execution and give me its FAQs as a table. " #### Prompts for the Databricks Field Guide server " Ask the Databricks Field Guide how to give each pull request its own Lakebase branch, and show me the CLI command it uses. " " What has Databricks changed in the last two weeks, according to the Field Guide? " " Find SQL examples of Unity Catalog row filters in the Field Guide, then check the docs.databricks.com links they cite for the current syntax. " " List the Field Guide's chapters, then read me the Lakebase chapter's section on why a branch costs almost nothing. " For TechFabric authors ### How to write a TechFabric blog post with your own agent Your agent writes the post from what you actually did, and the site checks it, builds the preview and leaves publishing to you. Writing needs a techfabric.com Microsoft account that is on the blog's author list. #### Set up the server Sign-in is accepted from Claude Code and Claude's custom connectors (Claude Desktop and claude.ai) only. Cursor, VS Code and other clients can read but can't sign in to write. In Claude Code, add the server once, for all your projects. ``` claude mcp add --transport http --scope user techfabric https://www.techfabric.com/mcp ``` Then run /mcp, pick techfabric and choose Authenticate. If you skip that, Claude Code asks you to sign in the first time your agent calls a write tool. In Claude Desktop or claude.ai, add a custom connector named techfabric with the URL above, as in how to connect , and choose Connect. #### Sign in with Microsoft Either way a browser opens on a techfabric.com page that names the client asking and where its access goes. Choose Continue with Microsoft and sign in with your @techfabric.com account. Access lasts an hour and renews itself for 30 days. #### The workflow, step by step Step 1 ##### Read the editorial rules Your agent reads the editorial_rules resource (techfabric://editorial/rules). It starts with client confidentiality, which decides what the post may say, then gives the frontmatter, this workflow and the house writing rules. Step 2 ##### Draft from your own work Your agent drafts the post, frontmatter and all, from what it knows about your own work, such as the repository you're in and your notes. The frontmatter is required, with a title and a description at least: check_draft, submit_draft and update_draft refuse a post with no frontmatter or no title. Step 3 ##### Check it and fix the findings Your agent runs check_draft, which runs the browser editor's own checks without filing a draft. A full run takes five to ten minutes. Each finding comes back with the exact passage it's about, so the agent fixes those passages and checks again. Step 4 ##### Submit it under your byline Your agent calls submit_draft. The post is filed under your own byline exactly as written. The site sets the author, today's date and draft: true, and adds the reading time if it's missing. It renders the diagrams, draws the cover and share image, builds the branded preview and emails it to you. The agent gets the draft's id back. Step 5 ##### Follow its status Your agent calls get_status with the draft's id, or subscribes to techfabric://drafts/{id}, for the check results, what's still failing, the preview, the editor link and the draft's base_sha. Step 6 ##### Revise with update_draft For revisions your agent calls update_draft with the id, the whole new post and that base_sha. A save commits the text and rebuilds the preview. It runs no checks. update_draft is refused if the draft has changed since that base_sha, or while it's open in the browser editor. Step 7 ##### Run checks and publish from the editor You read the preview, open the editor link and choose Run checks, or Run checks, then publish. Publish needs a check run that passed on exactly the saved text. A check_draft pass doesn't count, and any save after a run means running the checks again. The approve link in the preview email, when it has one, runs every check at approval instead and holds the post if one fails. #### What our side does and doesn't do Your text is the draft. Our blog writer doesn't redraft it and no model fact-checks it, so the facts are your agent's to check against the vendor's documentation before it submits. Where the editorial rules mention the blog writer's fact-check, that step doesn't run for your post. The site's checks still hold it, including the claims and known-errors checks, and publishing needs them to pass. Keep your edits in your agent and update_draft unless you want the writer involved. Replying to the preview email is different, because the reply goes to our blog writer. If the writer then needs a fact only you have, get_status asks you for it, or, in a client that supports MCP tasks, the open task does. #### Rules to know - No client names, data, metrics, schemas, incidents or code unless the client is approved for publication, and then only as far as the approval goes. The editorial rules list who's approved. When in doubt, generalise it or leave it out. - First-hand claims must be true. Write "we" only for what we did, and never let the agent invent a number, a result, a client or a quote. #### A prompt to start from Replace the placeholder with something you built or fixed. The prompt asks the agent to read the rules first. ``` Read techfabric.com's editorial rules (the techfabric editorial_rules resource), then write a TechFabric blog post about how we . Draft only from what I actually did and what's in this project; never name a client the rules don't approve, and ask me for any fact you don't have instead of inventing it. Run check_draft and fix every finding, then submit it with submit_draft and give me the preview and editor links. ``` #### Troubleshooting **Sign-in refused** The sign-in page says why. If it says your address is not on techfabric.com's author list yet, there's no byline to draft under, so ask Preetham Reddy to add you. If it says only TechFabric's Microsoft accounts or only @techfabric.com accounts can sign in, or that guest accounts can't, you signed in with another Microsoft account; sign in again with your own @techfabric.com one. If it says the app can't connect, the client isn't Claude Code or a Claude custom connector. **A check keeps failing** Open the editor link from get_status. Its checklist shows each failing item with the passage it's about. Fix it there, or have your agent fix it and call update_draft. **Too many calls** Signed in, the limits per hour are 600 calls, 20 new check runs, 6 submissions and 30 saves. Checking text that was already checked doesn't count as a new run. **Anything else** Ask Preetham Reddy, who runs the blog. Privacy ### What do TechFabric's MCP servers keep about you? Reading keeps nothing about what you ask. It needs no account and sets no cookie, and the only thing stored is a per-IP call count for the rate limit. Authors who sign in to write leave their drafts and address. Neither server keeps what you searched for or read. The log line for a call names the tool and the protocol revision, never its arguments. The only thing stored about an anonymous caller is a count of calls per IP address for the rate limit, which expires after about an hour. No read call reaches a language model on our side. Both servers answer from an index built when the site was deployed, and the Field Guide's server makes no outbound request at all. When a TechFabric author signs in to write, the server keeps the text they check, file or save, with their address against it, as the browser editor does. The privacy policy covers the rest of the site. FAQ ### MCP servers: common questions **What is an MCP server?** MCP, the Model Context Protocol, is an open standard for connecting AI assistants to other systems. An MCP server publishes a list of tools with their parameters, and an assistant such as Claude, Cursor or VS Code's agent mode calls them partway through a task and gets structured answers back. A remote server like ours is a single URL, so connecting one is a line of configuration and nothing to install. **Do I need an account to use TechFabric's MCP servers?** No. Every read tool on both servers works without signing in, and neither server sets a cookie. The only sign-in is on techfabric.com's write tools, which are for TechFabric's own blog authors and ask for a Microsoft account in the techfabric.com tenant. **Can an agent publish to techfabric.com through the MCP server?** No. Neither server has a publish tool. An author's agent can check a draft and file it under the author's own byline, and that draft then waits for the author, who reads the preview and publishes from the editor link like any other post. Anyone who isn't on the blog's author list can't reach the write tools at all. **Is the Field Guide server Databricks' own MCP server?** No. Databricks runs its own managed MCP servers inside a workspace, which act on your data and are documented on docs.databricks.com. The Field Guide server is built and run by TechFabric, a Databricks partner. It answers Databricks questions from our guide, with code and the documentation links each section cites, and it never connects to a workspace. **How current is the Databricks Field Guide server?** Its index is rebuilt on every deploy of the guide, so a new margin note or a rewritten chapter is in the server's answers as soon as that deploy is live. Each answer carries the commit that built it, when each page last changed, and how long a client may cache it. For exact syntax and limits, follow the docs.databricks.com links it returns, because Databricks' own documentation is the authority and the guide links to it instead of copying it. **What happens if my agent goes over the rate limit?** The server answers HTTP 429 with a Retry-After header, and the agent can try again after that. techfabric.com allows 300 anonymous calls per IP address per hour. The Field Guide allows 60 a minute and 300 an hour per IP address. **Are TechFabric's MCP servers in the official MCP Registry?** Both are. techfabric.com is listed as com.techfabric/techfabric and the Field Guide as com.techfabric.databricks/field-guide, and Glama lists both as connectors. A client that browses the registry finds them under those names, and the domains behind each name are proved by a key served from that domain.