Skip to content
Back to search

mcp-server-toolkit

Secure MCP server (TypeScript) with data tools for AI agents: read-only SQLite, CSV profiling, JMESPath, SSRF-safe fetch, diffs. Sandboxed, tested against injection-driven attacks.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/joaofrancowork--mcp-server-toolkit/badge.svg)](/listings/joaofrancowork--mcp-server-toolkit)

HTML

<a href="/listings/joaofrancowork--mcp-server-toolkit"><img src="/api/listings/joaofrancowork--mcp-server-toolkit/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Claude Desktop (`claude_desktop_config.json`) or any MCP host:

{
  "mcpServers": {
    "data": {
      "command": "npx",
      "args": ["-y", "github:joaofrancowork/mcp-server-toolkit", "--root", "/path/to/your/data"]
    }
  }
}

Claude Code: `claude mcp add data -- npx -y github:joaofrancowork/mcp-server-toolkit --root ./data`

From source:

git clone https://github.com/joaofrancowork/mcp-server-toolkit && cd mcp-server-toolkit
npm ci && npm run build
npm run inspect          # MCP Inspector against the example data

Docker (stdio): `docker run -i --rm -v "$PWD/data:/data:ro" mcp-server-toolkit`. Options: `--root <dir>` (or `MCP_TOOLKIT_ROOT`), `--no-fetch` to disable the only network tool, `--audit-log <file>` to append one JSON line per tool call (tool, arguments truncated to 200 characters, duration, error flag).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 120 lines
  1. # mcp-server-toolkit
  2. [![CI](https://github.com/joaofrancowork/mcp-server-toolkit/actions/workflows/ci.yml/badge.svg)](https://github.com/joaofrancowork/mcp-server-toolkit/actions/workflows/ci.yml)
  3. ![TypeScript](https://img.shields.io/badge/TypeScript-strict-black)
  4. ![Node](https://img.shields.io/badge/node-%E2%89%A522.13-black)
  5. ![License](https://img.shields.io/badge/license-MIT-black)
  6. A [Model Context Protocol](https://modelcontextprotocol.io) server that gives AI agents
  7. (Claude Desktop, Claude Code, any MCP client) **data tools**: read-only SQL, CSV profiling,
  8. JSON queries, web fetch and text diffs. It is built for the failure mode that matters with
  9. agents: **a prompt-injected model calling your tools**. Every tool is sandboxed, read-only and
  10. tested against the attacks.
  11. ![MCP Inspector running csv_profile against the example data](docs/inspector.png)
  12. ## Tools
  13. | Tool | What it does | Guardrails |
  14. |---|---|---|
  15. | `sqlite_schema` | Tables, columns and row counts of a SQLite file | Read-only connection |
  16. | `sqlite_query` | One SQL query, results as a Markdown table plus structured rows | Read-only connection **and** a single `SELECT`/`WITH`/`EXPLAIN`; `ATTACH`, `PRAGMA` and writes rejected; 500-row cap |
  17. | `csv_profile` | Row count, inferred types, nulls, uniques, ranges, top values, sample rows | Sandboxed path, 20 MB cap |
  18. | `json_query` | JMESPath over a JSON document | Input size cap; parse and expression errors returned to the model |
  19. | `text_diff` | Unified diff and word-level change counts | Input size cap |
  20. | `fetch_url` | Public web page as readable text | **SSRF protection** on every hop, http(s) only, no credentials in URLs, 2 MB cap, 10 s timeout, max 3 redirects |
  21. Plus a `toolkit://files` resource (what is available), a `toolkit://file/{path}` template, and
  22. an `explore_dataset` prompt that makes the model profile before it queries and show its
  23. evidence. All tools carry `readOnlyHint`; only `fetch_url` is `openWorldHint`.
  24. ## Quick start
  25. Claude Desktop (`claude_desktop_config.json`) or any MCP host:
  26. ```json
  27. {
  28. "mcpServers": {
  29. "data": {
  30. "command": "npx",
  31. "args": ["-y", "github:joaofrancowork/mcp-server-toolkit", "--root", "/path/to/your/data"]
  32. }
  33. }
  34. }
  35. ```
  36. Claude Code: `claude mcp add data -- npx -y github:joaofrancowork/mcp-server-toolkit --root ./data`
  37. From source:
  38. ```bash
  39. git clone https://github.com/joaofrancowork/mcp-server-toolkit && cd mcp-server-toolkit
  40. npm ci && npm run build
  41. npm run inspect # MCP Inspector against the example data
  42. ```
  43. Docker (stdio): `docker run -i --rm -v "$PWD/data:/data:ro" mcp-server-toolkit`.
  44. Options: `--root <dir>` (or `MCP_TOOLKIT_ROOT`), `--no-fetch` to disable the only network tool,
  45. `--audit-log <file>` to append one JSON line per tool call (tool, arguments truncated to 200
  46. characters, duration, error flag).
  47. ## Architecture
  48. ```mermaid
  49. flowchart LR
  50. H[MCP host<br/>Claude Desktop · Claude Code] <-->|JSON-RPC over stdio| S[McpServer]
  51. S --> T1[sqlite_*]
  52. S --> T2[csv_profile]
  53. S --> T3[json_query · text_diff]
  54. S --> T4[fetch_url]
  55. S --> R[resources · prompt]
  56. T1 & T2 & R --> SB{{Sandbox<br/>realpath under --root}}
  57. T1 --> RO{{read-only SQLite<br/>+ statement allow-list}}
  58. T4 --> SSRF{{DNS resolve every hop<br/>block private ranges}}
  59. SB --> FS[(your data dir)]
  60. ```
  61. ## Security model
  62. The model is not trusted: anything it reads (a web page, a CSV cell) may contain instructions
  63. from an attacker. The tools assume that and limit the blast radius.
  64. - **Filesystem.** Paths are resolved to their *real* path before the prefix check, so `../`,
  65. absolute paths and symlinks pointing outside `--root` are refused. Only expected extensions
  66. are opened.
  67. - **SQL.** Two independent layers: SQLite opens the file with `readOnly: true`, and the query
  68. must be one `SELECT`/`WITH`/`EXPLAIN` without write, `ATTACH` or `PRAGMA` keywords (string
  69. literals are ignored, so `WHERE name = 'drop'` still works).
  70. - **Network.** `fetch_url` resolves the host and rejects private, loopback, link-local, CGNAT,
  71. multicast and IPv4-mapped IPv6 addresses (cloud metadata at `169.254.169.254` included). It
  72. does this again for every redirect, which is how most SSRF filters are bypassed.
  73. - **Errors are results.** Expected failures come back as `isError` tool results with a readable
  74. message, so the model can correct itself instead of the session crashing.
  75. - **stdout is the protocol.** Logs go to stderr only.
  76. ## Testing
  77. ```bash
  78. npm test # 38 tests through a real MCP client over an in-memory transport, coverage gate
  79. npm run smoke # builds, spawns the server over stdio and calls a tool, as a host would
  80. ```
  81. The tests cover the protocol surface (tools, resources, prompt), the behaviour of each tool and
  82. the attacks: path traversal, symlink escape, SQL writes hidden in CTEs, multiple statements,
  83. `ATTACH`, SSRF through private IPs, DNS answers that mix public and private addresses,
  84. redirects to localhost, redirect loops and oversized responses.
  85. CI runs lint (Biome), `tsc --strict`, tests on Node 22 and 24, the stdio smoke test, a Docker
  86. build with an `initialize` handshake, and on a `v*` tag publishes a GitHub release with the
  87. packed tarball.
  88. ## Roadmap
  89. - [ ] Streamable HTTP transport with OAuth for remote deployments
  90. - [ ] Parquet support via DuckDB (read-only)
  91. - [ ] Per-tool rate limits and an audit log of tool calls
  92. - [ ] `fetch_url` content extraction with Readability
  93. ## License
  94. MIT. The example data is generated by `npm run examples` and is fictional.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.