mcp-forge
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-forge Add the Petstore OpenAPI and show me the tools"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-forge
Turn any docs URL into lean, callable MCP tools. No more pasting 1M-token API dumps into context.
Install
npx mcp-forgeRequires Node >= 20.
Related MCP server: DocsFeed MCP Server
Claude config
Add to your Claude Desktop / Claude Code MCP config:
{
"mcpServers": {
"forge": {
"command": "node",
"args": ["dist/index.js"]
}
}
}If installed globally, point args at your install path (e.g. /path/to/mcp-forge/dist/index.js). Then restart Claude.
30-sec demo
Point Forge at an OpenAPI URL, list what it forged, then call one:
forge_add({ "sourceUrl": "https://petstore3.swagger.io/api/v3/openapi.json" })
// -> { "tools": ["listPets", "getPetById", ...] }
forge_list()
// -> [{ "name": "listPets", "tokens": 312 }, ...]
forge_call({ "toolName": "getPetById", "args": { "petId": 1 } })
// -> { "status": 200, "body": "{...}", "truncated": false }Works with OpenAPI / Swagger JSON, llms.txt, and plain HTML docs pages. Forged tools are saved to a local store and executed with timeout + truncation guards.
Why: token table
Blog-era bloat vs. forged tools (per widely cited ecosystem reports — your mileage varies by API):
Source | Tokens in context |
Forge (per forged tool) | < 3K |
GitHub API surface | ~45K |
Cloudflare API surface | ~1.17M |
Forge fetches the docs once, extracts lean tool definitions (method + path + args), and only the small definition enters context — never the full spec.
Token figures cite known ecosystem reports about MCP server token costs, not benchmarks run by this repo. No fake claims: measure your own API before quoting numbers.
Security notes
SSRF block: private/loopback/link-local hosts and non-http(s) schemes are rejected before fetch.
15s timeout: both docs fetch and forged-tool execution abort after 15 seconds (
AbortController).10KB truncate: execution response bodies are cut at 10,240 chars with
truncated: true— context never fills with a giant dump.No secrets stored: the local store holds only tool definitions (name, method, path, source URL). No API keys, tokens, or response bodies are persisted.
License
MIT.
Available Tools
3 toolsforge_addA
Fetch a docs URL (OpenAPI / llms.txt / HTML) and forge lean callable tools from it. Returns {tools: names}.
Inputs: sourceUrl (required).
forge_callB
Call a forged tool by name with args. Loads tool JSON from store then executes with timeout + truncation.
Inputs: toolName (required), args.
forge_listA
List forged tools in the local store.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a clearly distinct role: forge_add creates tools from a URL, forge_list enumerates them, and forge_call executes them. There is no overlap or ambiguity between the three operations.
All tools share the forge_ prefix followed by a lowercase imperative verb: add, list, call. The naming is completely consistent and predictable.
Three tools is appropriate for this focused server's scope: one to create tools, one to list them, and one to call them. Each tool earns its place and no extra tool is needed for the core workflow.
The add/list/call workflow covers the essential lifecycle of dynamically forged tools. However, there is no delete or refresh operation, so stale or outdated forged tools cannot be removed.
Maintenance
Related MCP Connectors
Turn a GitHub repo or docs site into agent-ready context: pack it or search it, over MCP.
Query any docs site via MCP. Submit a URL, ask questions, get cited answers.
MCP tools for AI agents: render URLs to image/PDF, check link health, convert HTML/CSV/JSON.
The OpenRouter for tools. One MCP connection gives any AI agent 254 hosted tools, pay per call.
Related MCP Servers
- AlicenseNot gradedqualityNot gradedmaintenanceEnables remote MCP access to tools for URL-to-markdown conversion, web and image search, academic search, embeddings, reranking, classification, deduplication, and PDF extraction.Apache 2.0
- FlicenseNot gradedqualityBmaintenanceEnables users to ingest documentation from URLs or sitemaps, embed it into a searchable vector index, and expose authenticated MCP tools so AI clients can search and retrieve documentation content directly.-
- AlicenseAqualityCmaintenanceProvides MCP tools for checking URL accessibility, evaluating whether webpage content is recently updated, and updating markdown tables keyed by URL, letting AI agents offload deterministic tasks to code and reduce token usage.3MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to read public web content as clean Markdown, search a private library of previously read documents, and access these tools via MCP, REST API, CLI, and WebMCP.4 npm22MIT