mcpField
A
AgentToolbox Docs Pack

AgentToolbox Docs Pack

Remote
github.com/agenttoolbox2026Developer Tools0Updated 3w ago
Repository

About AgentToolbox Docs Pack

AgentToolbox Docs Pack is an experimental service for agents and developers that returns exact excerpts from public documentation matching literal query terms. Each successful paid pack covers up to five URLs and includes source hashes, offsets, and matched headings.

Key features

  1. 01Exact excerpts matched to query terms
  2. 02Support for up to five documentation URLs
  3. 03Source hashes, offsets, and matched headings
  4. 04Free fixed-input example
  5. 05Pay per successful pack using x402

Use cases

  1. 01Find relevant passages across public documentation
  2. 02Retrieve bounded excerpts for an agent workflow

MCP support

The service provides MCP discovery at `/mcp` and lists the Docs Pack in the MCP Registry. Its paid invocation is also available over HTTP.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://agnttoolbx.agenttoolbox2026.workers.dev/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

Details

From the project README

AgentToolbox

Agent-first home: https://agnttoolbx.agenttoolbox2026.workers.dev

Docs Pack is the first experimental product: up to five supported public documentation URLs and literal query terms return bounded exact excerpts, source hashes, offsets and matched headings. Price: $0.01 USDC on Base per successful whole pack, using HTTP x402 v2. No source match or failed source means no settlement. Live payment verification remains user-controlled and has not been performed by the implementation agent.

Try it now

Free example — no wallet, no payment, fixed public inputs:

curl -s "https://agnttoolbx.agenttoolbox2026.workers.dev/v1/products/docs-pack/example" | python3 -m json.tool | head -60

Paid invoke — returns HTTP 402 with the x402 challenge; settle it with any x402 v2-capable wallet/client and resubmit:

IDEMPOTENCY_KEY="$(python3 -c 'import uuid; print(uuid.uuid4().hex + uuid.uuid4().hex[:32])')"
curl -s -X POST "https://agnttoolbx.agenttoolbox2026.workers.dev/v1/products/docs-pack/invoke" \
  -H 'Content-Type: application/json' \
  -H "Idempotency-Key: $IDEMPOTENCY_KEY" \
  -d '{"version":"0.1.0","input":{"urls":["https://docs.python.org/3/library/asyncio.html"],"query":"event loop","max_excerpt_chars":4000},"max_charge_usdc_atomic":10000}'

Runnable scripts: examples/free-call.sh, examples/paid-call.sh.

Endpoints

  • Contract: /v1/products/docs-pack
  • Real, fixed free example: /v1/products/docs-pack/example
  • Paid invocation: POST /v1/products/docs-pack/invoke
  • HTTP discovery: /v1/products; MCP discovery: /mcp
  • MCP Registry: io.github.agenttoolbox2026/docs-pack (v0.1.0)
  • Machine instructions: /llms.txt; schemas: /openapi.json
  • Human observer page: /humans; public paid-purchase total: /v1/stats

The human counter counts completed live paid purchases, including repeats. It excludes synthetic tests, mocks, failed/pending operations, replays and self-purchases from the receiving wallet. It does not identify unique agents. Private operations and reports remain authenticated. The original retry gate is retired; its source, D1 database and history remain preserved.

Run

Node.js 24+ and pnpm 11.19.0:

pnpm install --frozen-lockfile
pnpm --dir platform install --frozen-lockfile
pnpm dev
pnpm test
pnpm check
pnpm worker:check
pnpm smoke https://agnttoolbx.agenttoolbox2026.workers.dev
pnpm report:remote

Platform code is in platform/; root src/, test/ and scripts/ preserve the legacy experiment. The original icon is unchanged. No paid Cloudflare resource, facilitator account, top-up, wallet secret or automatic billing was added.

Read the brief, product and benchmark, operations and payment limits.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.