M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
O

Open Agent Builder

Community
Not a standard server
Visual workflow builder for AI agents powered by Firecrawl

This is not a typical connect-and-use MCP server; read the source repository before using it.

How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #422 of 438
Stars
★ 2.6k Popular
Transport
stdio (local process)
Runtime
Node.js 18+
Credentials
API key / credential required
Last commit
⚠
27FMRS · D

Open Agent Builder is a rich visual workflow builder leveraging Firecrawl for web scraping, with MCP protocol support for Anthropic Claude. It offers an intuitive drag-and-drop interface and real-time execution. However, it requires setup with multiple external services (Convex, Clerk, Firecrawl), and MCP support for OpenAI and Groq is not yet available. Suitable for users wanting to build AI agent workflows and willing to handle complex deployment.

Strongest · Documentation 8/20 Weakest · Reliability 2/20

Reliability
2/20
Security and permissions
6/20
Maintenance
7/20
Documentation
8/20
Setup experience
4/20
Why each score
Reliability 2/20
The README describes a Next.js visual workflow application rather than a standalone MCP server; no MCP server manifest, tool list, or handshake details are supplied, and there is no test or CI evidence. Startup, MCP handshake, and tool behavior cannot be verified, so a conservative score of 2 is given.
Security and permissions 6/20
Clerk authentication, user-level API keys, and a User Approval node are present, with no sign of malware or hardcoded real secrets. However, least privilege, encryption of stored keys, scoping/confirmation of MCP operations, and full data-flow disclosure are missing. Main risks are visible but incomplete; score 6.
Maintenance 7/20
The README shows an active development roadmap, and GitHub metadata indicates the repository is not archived with 14 open issues. There is no release history, dependency update policy, or security response channel, and the license is only claimed via a README badge while the metadata says unknown. Governance and versioning gaps justify a score of 7.
Documentation 8/20
Installation, authentication, environment variables, example workflows, and architecture diagrams are reasonably complete for the web application. Missing MCP server tool parameters, limits, cost, and troubleshooting details, and the README does not fully match an MCP server identity, so documentation scores 8.
Setup experience 4/20
Setup requires Node.js, Firecrawl, Convex, Clerk, and many environment variables, with no MCP client connection examples. An install path exists for the web app, but using it as an MCP server is complex and fragile, hence a score of 4.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessRuns commands or codeUses the networkConnects to a database

Best for

  • Developers and teams wanting to visually build AI agent workflows
  • Users needing integration with Firecrawl for web scraping
  • Users wanting to use Anthropic Claude's MCP tool support
  • Workflows requiring real-time execution and streaming updates

Not for

  • Users needing simple scripts without a visual interface
  • Users needing full MCP support with OpenAI or Groq (currently limited)
  • Users wanting to deploy as a standalone MCP server rather than an application

Required permissions

  • Requires Firecrawl API key and network access for scraping
  • Requires LLM API keys (e.g., Anthropic, OpenAI, Groq)
  • Requires Convex database and Clerk authentication for deployment
  • Optional E2B key for sandboxed code execution

Risks and side effects

  • Project is actively developed; some features may be unstable
  • Dependency on external services (Firecrawl, Convex, Clerk) costs and availability
  • Exposure of LLM and Firecrawl API keys; must be managed securely
  • Supply chain risk: relies on many third-party npm packages; review dependencies

Setup

Before you start

Runtime：Node.js 18+

NEXT_PUBLIC_CONVEX_URL required Convex database URL, generated automatically when you run npx convex dev to initialize your Convex project.
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY required Clerk public key from the API Keys page of your Clerk dashboard.
CLERK_SECRET_KEY requiredsecret Clerk secret key from the API Keys page of your Clerk dashboard.
CLERK_JWT_ISSUER_DOMAIN required Clerk JWT issuer URL, copied from JWT Templates → Convex in the Clerk dashboard.
FIRECRAWL_API_KEY requiredsecret Firecrawl web scraping API key (required), obtained at firecrawl.dev.
ANTHROPIC_API_KEY optionalsecret Optional default LLM key, recommended for MCP support, from anthropic.com.
OPENAI_API_KEY optionalsecret Optional default LLM key (MCP support in development), from platform.openai.com.
GROQ_API_KEY optionalsecret Optional default LLM key (MCP support in development), from groq.com.
E2B_API_KEY optionalsecret Optional, enables sandboxed code execution in Transform nodes, from e2b.dev.
  1. Clone repo: git clone https://github.com/firecrawl/open-agent-builder.git && cd open-agent-builder && npm install
  2. Set up Convex: npm install -g convex && npx convex dev
  3. Set up Clerk: create app at clerk.com, configure API keys and JWT template
  4. Configure Convex auth: edit convex/auth.config.ts and run npx convex dev
  5. Set up Firecrawl: get API key and add to .env.local as FIRECRAWL_API_KEY
  6. Optionally set default LLM provider (e.g., ANTHROPIC_API_KEY)
  7. Optionally set E2B key
  8. Run: terminal 1: npx convex dev, terminal 2: npm run dev, or use npm run dev:all

Check that it works

Start the app, visit http://localhost:3000, sign up, add an LLM key in Settings → API Keys, then run the 'Simple Web Scraper' template with a URL; a streaming execution returning a scraped summary confirms setup. Note: this is a visual workflow builder whose Agent nodes can call Firecrawl MCP tools (scrape, search, crawl), not a standalone MCP server to connect a client to.

Troubleshooting

  1. Ensure all required environment variables are set correctly
  2. Ensure Convex dev server is running in a separate terminal
  3. Check Firecrawl API key validity and quota
  4. For MCP tools, use Anthropic Claude as default provider
  5. Verify Clerk configuration in .env.local, especially JWT issuer domain
  6. When running npx convex dev, ensure auth config is pushed correctly

Things to try

Once connected, you can ask your AI assistant things like:

  • Scrape https://firecrawl.dev and summarize the page with AI
  • Search the web for a topic, scrape the top results, and synthesize a research brief
  • Monitor a product page and extract price changes
  • Add a human approval step before publishing content

Use cases

Building web scraping and data extraction workflows
Creating multi-step AI agent pipelines
Automating research and content generation
Performing data transformation and analysis
Implementing web automation with human approvals

Overview

Open Agent Builder is a visual workflow builder for creating AI agent pipelines powered by Firecrawl. Its drag-and-drop interface helps design complex agent workflows with real-time streaming execution. It includes 8 core node types, LangGraph execution engine, Clerk authentication, Convex database, and MCP protocol support for tool integration.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision be856e57f812 Data synced 2026-10-11 Read the FMRS scoring method