← Blog

Pipeworx now implements the full MCP protocol

Tools, Resources, Prompts, plus five meta-tools that collapse multi-call patterns into one. What most MCP servers ship versus what an agent actually needs.

Most MCP servers implement one slice of the protocol: tool calling. The MCP spec has three nouns — Tools (do things), Resources (read citable entities), Prompts (workflow templates) — and the typical server only handles the first.

Pipeworx now ships all three, plus five gateway-native meta-tools that collapse common multi-call patterns into a single call.

If you connect an agent to https://gateway.pipeworx.io/mcp today, it sees:

  • 1,007 tools across 295 packs (SEC EDGAR, FDA, FRED, Census, EPA, ClinicalTrials, ATTOM, RxNorm, USPTO, …)
  • Each tool’s inputSchema includes concrete examples — first-call argument shape comes for free.
  • Each tool also includes an outputSchema describing the response — agents can plan downstream operations before calling.
  • 15 Resource templates for citable entities (pipeworx://edgar/company/{cik}/filings, pipeworx://clinicaltrials/study/{nct_id}, pipeworx://rxnorm/concept/{rxcui}, …)
  • 12 Prompt playbooks for common workflows (research_public_company, drug_safety_profile, trade_exposure, housing_market_brief, regulatory_landscape, financial_health_check, …)
  • serverInstructions in the initialize response — a cold-start briefing telling agents how to use the server effectively.

The five meta-tools

These collapse what would otherwise be 5–10 sequential agent calls.

ask_pipeworx(question)

Plain-English routing. Picks the right tool from 1,000+, fills the arguments, returns the answer.

discover_tools(task)

Semantic search over the catalog. Returns the 20 most relevant tools for a task — narrows the search space when the agent doesn’t know the right tool yet.

resolve_entity({type, value})

Cross-pack identity resolution.

resolve_entity({ type: "company", value: "AAPL" })
// → { resolved: true,
//     ids: { cik: "0000320193", ticker: "AAPL", company_name: "Apple Inc." },
//     resources: {
//       edgar_filings: "pipeworx://edgar/company/0000320193/filings",
//       edgar_facts:   "pipeworx://edgar/company/0000320193/facts" } }

resolve_entity({ type: "drug", value: "ozempic" })
// → { resolved: true,
//     ids: { rxcui: "1991311", product_name: "...semaglutide... [Ozempic]" },
//     resources: { rxnorm_concept: "pipeworx://rxnorm/concept/1991311", ... } }

compare_entities({type, values})

Side-by-side in parallel.

compare_entities({ type: "company", values: ["AAPL", "MSFT"] })
// → revenue / net income / cash / long-term debt for each, plus deltas
//   pivoted by concept. ~10 sequential calls collapsed into one parallel call.

compare_entities({ type: "drug", values: ["ozempic", "mounjaro"] })
// → ozempic:  63,000 adverse events | 2 approvals | 104 trials
//   mounjaro: 87,308 adverse events | 1 approval  | 34 trials

remember / recall / forget

Persistent agent memory across sessions.

Why this matters

Most agentic flows look like: connect, list tools, scan descriptions, call the right one, parse the response, decide next step, call another, … Each round trip is latency, context window, and an opportunity to get something wrong.

Pipeworx pushes the cost of “figure out what to call” and “stitch the results together” off the agent and into the gateway:

  • tools/list carries everything an agent needs to make the first call correctly.
  • Errors include _meta.examples and _meta.alternatives so retries self-correct.
  • Resources let agents cite entities by URI so output is durable and shareable.
  • Prompts encode the optimal multi-step plan for common domain tasks.
  • Meta-tools collapse the “I need 5 things about X” pattern into one call.

The result: fewer round trips, fewer context tokens, fewer chances to drift, more stable citations in the agent’s output.

Try it

Add Pipeworx to any MCP client:

{
  "mcpServers": {
    "pipeworx": { "url": "https://gateway.pipeworx.io/mcp" }
  }
}

50 calls/day anonymous, 2,000/day with a free account at pipeworx.io, unlimited on paid.

The gateway’s initialize response now includes a server-side briefing telling your agent about the meta-tools, prompts, and resource URIs — most clients surface this automatically. Or read the llms.txt for the full picture.

If you build agents and want a single URL that gives you primary-source live data with full MCP semantics, this is for you.