context7 (MCP retrieval) ​

context7 is an Upstash MCP service that crawls a repo's documentation and serves it to any agent that speaks MCP. Register chartjs2img once; every MCP-capable agent (Claude Code, Cursor, Gemini CLI, Codex, anything using the spec) can resolve + query the docs without installing anything else.

This is a read-only path — context7 doesn't give agents the ability to render charts. For that, combine it with the Claude plugin or gh skill tutorial.

What context7 does for chartjs2img ​

When an agent asks "how do I show data labels on a bar chart in chartjs2img?", context7:

1. Resolves chartjs2img to our registered library ID.
2. Searches our docs for the relevant passage (datalabels section + the bar example + plugins/datalabels.display: true requirement).
3. Returns it to the agent as MCP tool output.

No CLI install, no skill install, no binary download. Pure docs retrieval.

How it's set up ​

A context7.json file at the repo root tells context7 what to index:

{
  "$schema": "https://context7.com/schema/context7.json",
  "projectTitle": "chartjs2img",
  "description": "Server-side Chart.js rendering service...",
  "folders": ["docs/en", "examples", "src/llm-docs"],
  "excludeFolders": ["docs/.vitepress", "docs/public", "docs/ja", "node_modules", "dist", "tests"],
  "rules": ["Input is always Chart.js configuration JSON. ...", "..."]
}

The rules array contains 9 chartjs2img-specific pitfalls — JSON input shape, no-functions constraint, animation forced off, datalabels off-by-default, X-Chart-Messages feedback, exit codes, Chromium availability. Agents always see these, regardless of which specific doc passage they retrieve.

The live file: github.com/ideamans/chartjs2img/blob/main/context7.json.

How to use it from an MCP-capable agent ​

Claude Code ​

In a session, two MCP tools become relevant:

• mcp__context7__resolve-library-id — input: "chartjs2img" or a vague phrase; output: the canonical library id.
• mcp__context7__query-docs — input: library id + a natural-language question; output: the retrieved passages.

Example flow the agent runs automatically:

user: "Draw a pie chart with data labels."
assistant calls mcp__context7__resolve-library-id("chartjs2img")
      → returns "/ideamans/chartjs2img"
assistant calls mcp__context7__query-docs(
      libraryId="/ideamans/chartjs2img",
      query="pie chart with data labels")
      → returns the pie example + the datalabels section
assistant drafts a JSON config and (if the plugin is installed)
          calls /chartjs2img-render to produce the PNG.

Other MCP hosts ​

Any agent host that speaks MCP can add context7 as a server:

# In your MCP host's config
servers:
  context7:
    url: https://mcp.context7.com/

Check the host's docs for the exact configuration path (each host stores MCP server configs differently).

When context7 is helpful vs not ​

• Helpful when you need answers to "what does option X do" / "what plugin handles feature Y" / "what's the shape of data for chart type Z". One-shot retrieval, no install.
• Not helpful for actually rendering — that requires the CLI binary. Use Claude plugin or gh skill for the render-loop.
• Complementary: context7 + the author/render skills is the best combo. context7 surfaces the right option shape; the render skill validates it produces a clean PNG.

Checking that the registration succeeds ​

After we push context7.json, context7 crawls on its own schedule (typically a few hours). From the context7 web UI (context7.com/add-package) you can request an expedited crawl.

Once indexed, test from Claude Code:

mcp__context7__resolve-library-id  → ask for "chartjs2img"

Should return something like /ideamans/chartjs2img. If it returns empty, the crawl hasn't completed yet; check back in an hour or two.

What we exclude from the index ​

• docs/ja/** — Japanese is a mirror of English, and duplicated content hurts retrieval quality. English is treated as canonical.
• docs/.vitepress/** — build config, not content.
• docs/public/** — generated artifacts (llms.txt, example PNGs).
• node_modules/, dist/, tests/ — not docs.

If you need Japanese retrieval, file an issue — we can add a second context7 registration for the JA docs.

See also ​

• llms.txt — a parallel, simpler discovery file at the site root.
• CLI reference — chartjs2img llm produces the same reference offline.
• context7 docs — how to register your own projects.