Install
mcp_config.json
{
"mcpServers": {
"com-datpaq-datpaq": {
"url": "https://mcp.datpaq.com/mcp",
"type": "streamable-http"
}
}
}Documentation
Datpaq MCP Server
Hosted Model Context Protocol server for the Datpaq API.
Connect · Configure your client · Self-host · Dashboard ↗
Hosted at https://mcp.datpaq.com/ — also installable locally and self-hostable.
datpaq-mcp-http exposes the Datpaq API as MCP tools over streamable HTTP, so any MCP-compatible client (Claude Desktop, Cursor, Codex, Cline, Continue, agent frameworks) can call Datpaq endpoints with a single connection.
For the local CLI (and the stdio MCP that ships with it for Claude Desktop), see github.com/datpaq/cli.
Connect
| URL | https://mcp.datpaq.com/ |
| Transport | streamable HTTP |
| Protocol | 2026-07-28 and 2025-11-25 (and earlier) — see Protocol versions |
| Auth | Authorization: Bearer YOUR_DATPAQ_API_KEY (per request) |
| Tools | Hosted tools across 35 active APIs |
Verify with curl — modern (2026-07-28):
curl -s -X POST https://mcp.datpaq.com/ -H "Authorization: Bearer $YOUR_DATPAQ_API_KEY" -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "MCP-Protocol-Version: 2026-07-28" -H "Mcp-Method: server/discover" -d '{"jsonrpc":"2.0","id":1,"method":"server/discover","params":{"_meta":{"io.modelcontextprotocol/protocolVersion":"2026-07-28","io.modelcontextprotocol/clientInfo":{"name":"curl","version":"0"},"io.modelcontextprotocol/clientCapabilities":{}}}}'
Or legacy (initialize handshake), which keeps working unchanged:
curl -s -X POST https://mcp.datpaq.com/ -H "Authorization: Bearer $YOUR_DATPAQ_API_KEY" -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"curl","version":"0"}}}'
Protocol versions
The server is dual-era: it speaks both the stateless 2026-07-28 revision and the
handshake-based revisions (2025-11-25 and earlier) on the same endpoint, and picks per
request. You do not need to configure anything — point your client at the URL and it will
use whichever it supports.
| Era | Versions | How it's served |
|---|---|---|
| Modern | 2026-07-28 | Stateless. Every request carries its version, client info, and capabilities in _meta, mirrored into the MCP-Protocol-Version / Mcp-Method / Mcp-Name headers. Implements server/discover, tools/list, tools/call. |
| Legacy | 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05 | The initialize handshake, exactly as before. Sessions, GET SSE streams, and Mcp-Session-Id all still work. |
Notes for the modern era:
tools/listandserver/discoverreturnttlMsandcacheScope: "public"caching hints. The tool catalog is identical for every tenant and only changes on redeploy, so clients can cache it and skip re-fetching every tool definition on each reconnect.Mcp-MethodandMcp-Nameare validated against the request body. A header that disagrees with the body is rejected with-32020 HeaderMismatch; a missing header is served anyway, so clients that have not implemented header mirroring yet still work.- An unsupported version gets
400+-32022listing the versions we do support, so a client can retry rather than fail. - Features the spec deprecated in
2026-07-28— roots, sampling, logging, DCR, and the 2024-11-05 HTTP+SSE transport — were never used by this server, so nothing changes for you there.
Auth model: every request must present a Datpaq API key as Authorization: Bearer <key> — the server holds no credentials of its own, so a single instance can serve many tenants, each billed against their own Datpaq account.
Configure your client
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"datpaq": {
"url": "https://mcp.datpaq.com/",
"headers": {
"Authorization": "Bearer YOUR_DATPAQ_API_KEY"
}
}
}
}
Cursor (~/.cursor/mcp.json or project .cursor/mcp.json):
{
"mcpServers": {
"datpaq": {
"url": "https://mcp.datpaq.com/",
"headers": {
"Authorization": "Bearer YOUR_DATPAQ_API_KEY"
}
}
}
}
Codex CLI:
export DATPAQ_API_KEY="YOUR_DATPAQ_API_KEY"
codex mcp add datpaq --url https://mcp.datpaq.com/ --bearer-token-env-var DATPAQ_API_KEY
Codex TOML:
[mcp_servers.datpaq]
url = "https://mcp.datpaq.com/"
bearer_token_env_var = "YOUR_DATPAQ_API_KEY"
For full step-by-step instructions per client (including Cline and Continue), see the Datpaq MCP docs.
Self-host
Install via Go:
go install github.com/datpaq/mcp/cmd/datpaq-mcp-http@latest
Requires Go 1.26.3+.
Or pull the Docker image:
docker build -t datpaq-mcp-http -f cmd/datpaq-mcp-http/Dockerfile .
docker run -p 8080:8080 datpaq-mcp-http
Run it:
datpaq-mcp-http --addr :8080
Then point an MCP client at http://localhost:8080/.
Ships with a Fly.io config for one-command deploys:
fly launch --config fly.toml
fly deploy
No secrets to configure — clients authenticate per request.
What's exposed
The server registers one MCP tool per active Datpaq API endpoint. The active set is curated in internal/cli/active-apis.json — add a slug there when a new API ships on datpaq.com, redeploy, and the tool appears for every connected client.
Inactive APIs are scrubbed entirely from the MCP surface (unlike the CLI, which still lets you invoke them directly). This keeps the hosted tool list focused on production-ready endpoints.
Tools that depend on local state or shell-out to the user's machine (auth flows, config writers, etc.) are deliberately not registered — see internal/mcp/public_tools.go.
Configuration
| Variable | Purpose |
|---|---|
DATPAQ_BASE_URL | Override the API base URL (default: https://datpaq.com/api/v1) — handy for pointing at staging |
PORT | HTTP listen port (default: 8080, also configurable via --addr) |
The Datpaq API key is not a server-side variable. Each request must include it as Authorization: Bearer <key>. See the Quickstart auth model.
Managing active APIs
The active set is curated in internal/cli/active-apis.json. The CLI repo is the canonical place to regenerate that manifest from the admin dashboard export (or, in Phase 2, GET /api/v1/catalog/active). This repo receives updates via sync — it does not ship its own fetch tooling.
{
"active": [
"convert-time",
"ip-geolocation",
"..."
]
}
Canonical workflow (sibling checkout at ../CLI):
# In the CLI repo
cd ../CLI
make fetch-active-apis CATALOG_INPUT=./export.json # or CATALOG_URL=... after Phase 2
make sync-active-apis # CLI → MCP
# Then here
make build && make test
# commit internal/cli/active-apis.json + code changes, deploy
If you edited active-apis.json in this repo by mistake, push it back to CLI with ./scripts/sync-active-apis.sh (MCP → CLI only). Prefer the CLI-first flow above for normal updates.
Development
git clone https://github.com/datpaq/mcp && cd mcp
make build
./bin/datpaq-mcp-http --addr :8080
Run tests:
make test
License
Apache 2.0 — see LICENSE.
Sourced from the repository README.
More in Developer Tools
- N8nFair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.202,280
- Gemini CliAn open-source AI agent that brings the power of Gemini directly into your terminal.106,664
- World MonitorLive global intelligence: real-time markets, conflicts, country risk, chokepoints, energy. 39 tools.84,026
- WorldmonitorReal-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface84,024
- Scrapling🕷️ An adaptive Web Scraping framework that handles everything from a single request to a full-scale crawl!76,275
- Ruflo🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated69,270