Agentskb

io.github.CranotPublished by GitHub account Cranotv0.1.1Updated Oct 24, 2026

Researched knowledge for AI agents. Canonical answers from official sources.

VerifiedSTDIODesktop onlyKnowledge & MemoryDeveloper Tools

Overview

AI-generated overview

AgentsKB gives an assistant researched, source-backed answers from official documentation, specs and RFCs, plus tools to browse its Q&A knowledge base.

What you can do
  • Ask a natural-language question and get an expert-style answer with sources, optionally scoped to a domain and result limit
  • Check what verified knowledge already exists for a task, and list foundational prerequisite topics for a domain
  • Browse and search the Q&A knowledge base by topic, and navigate its domain taxonomy
  • Read coverage statistics and check whether a strict or pinned-answer mode is active
  • Expose two resources describing the detected project stack and a matching system-prompt fragment
When to use it
Useful when you want the assistant to answer technical questions from authoritative sources rather than guess, especially around databases, ORMs, web frameworks and infrastructure tooling. Also handy for checking whether a topic is covered before relying on it.
Requirements
A remote MCP endpoint over streamable HTTP; no API key or signup is required. Clients that only launch local processes need a bridge such as mcp-remote via npx. A plain HTTP endpoint is also offered for non-MCP use.
Before you install
Questions you ask are sent to a third-party hosted service, so avoid including secrets, credentials or private code. Coverage is uneven; uncovered topics are queued for research and return a queued notice instead of an answer. The npm CLI package is a separate tool and is not needed for the endpoint.

Installation

In SourceWeft

  1. Open Agentskb in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

AgentsKB MCP Server

Give Your AI Expert-Level Knowledge

Researched answers to any problem. Your AI, upgraded.

About this repository: AgentsKB is a hosted MCP server — there is no server to build or run locally, and this repository contains no source code. It exists to document the endpoint and to collect bug reports. Point your MCP client at the URL below and you are done.

Installation

The server is hosted at https://mcp.agentskb.com/mcp-kb and speaks MCP over streamable HTTP. No API key and no signup are required.

Claude Code

bash
claude mcp add --transport http agentskb https://mcp.agentskb.com/mcp-kb

Add --scope user to make it available in every project instead of just the current one. Verify the connection with:

bash
claude mcp list

Claude Desktop, Cursor, Cline, and other MCP clients

Add this to your client's MCP configuration file:

json
{  "mcpServers": {    "agentskb": {      "type": "http",      "url": "https://mcp.agentskb.com/mcp-kb"    }  }}

Config file locations:

ClientFile
Claude Code (per-project).mcp.json in the project root
Claude Code (global)~/.claude.json
Claude Desktopclaude_desktop_config.json
Cursor (global)~/.cursor/mcp.json
Cursor (per-project).cursor/mcp.json in the project root

Cursor does not use the type field — for Cursor, list url on its own.

Some older clients do not support remote HTTP servers and only launch local processes. If yours is one of them, bridge to the endpoint with mcp-remote:

json
{  "mcpServers": {    "agentskb": {      "command": "npx",      "args": ["-y", "mcp-remote", "https://mcp.agentskb.com/mcp-kb"]    }  }}

REST API (no MCP client)

The knowledge base is also reachable over plain HTTP, with no key:

bash
curl -X POST https://agentskb-api.agentskb.com/api/free/ask \  -H "Content-Type: application/json" \  -d '{"question":"What is a PostgreSQL index scan?"}'

What makes us different

Your AI searches. We research. That's the difference.

  • Dedicated research per question - not a side task while debugging
  • Official sources only - docs, specs, RFCs
  • Would rather not answer than guess wrong

Tools

ToolArgumentsDescription
ask_questionquestion*, domain, limitExpert-quality answers with sources
preflight_checktask*, depthSee what verified knowledge exists for your task
search_questionsquery*, domain, limitBrowse Q&As by topic
get_stats—Coverage metrics
consult_taxonomypath, depthNavigate the knowledge taxonomy
check_prerequisitesdomain*Foundational topics to understand first
get_lockfile_status—Check whether strict/pinned-answer mode is active

* required. question and query expect a full natural-language question — bare keywords such as mysql are rejected.

The server also exposes two resources: agentskb://context/current (detected project stack) and agentskb://context/prompt (a system-prompt fragment for that stack).

Coverage

55,000+ Q&As across 287 domains. The deepest coverage today is:

  • Databases and ORMs: PostgreSQL, Prisma, Drizzle
  • Frameworks: Next.js, NextAuth, Vue
  • Infrastructure and tooling: Docker, Bash, nginx, Git

Call get_stats for live counts, or consult_taxonomy to browse the full domain list.

Coverage is uneven and still growing. If you ask about a topic that is not yet covered, the server queues it for research and replies Topic '<name>' research queued instead of returning an answer — ask again later.

Reporting bugs

Open an issue on this repository: https://github.com/Cranot/agentskb-mcp/issues

Related

License

MIT

Source: README.md at commit 879aa77

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v0.1.1LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.