ByteAsk Embedded MCP

Community

Page-cited evidence retrieval for embedded and firmware coding agents.

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #199 of 438
Stars: ★ 25 Popular
Transport: stdio (local process) · Streamable HTTP
Runtime: Hosted (no local runtime) · Python 3.10+
Credentials: No credential needed
License: MIT
Last commit: 4 months ago
Tools: 3

54FMRS · D

ByteAsk Embedded MCP offers a unique and rigorous retrieval approach, focusing on exact facts for embedded development with cited outputs, suitable for high-fidelity agents. However, users should be aware of the distinction between hosted and self-hosted capabilities and the authentication limitations.

Strongest · Documentation 14/20 Weakest · Security and permissions 8/20

Reliability

10/20

Security and permissions

8/20

Maintenance

9/20

Documentation

14/20

Setup experience

13/20

Why each score

Reliability 10/20: README documents toolset, layout, and offline pytest suite, but no verifiable CI workflow or committed test contents are supplied, so handshake and tool behavior cannot be statically confirmed; the real retrieval engine lives outside the repo behind a seam (SampleBackend is demo-only), so key paths are not statically reproducible. Deductions: no execution evidence, closed-core implementation.
Security and permissions 8/20: No malware indicators; the repo candidly discloses the bearer check as a stub, unauthenticated-by-default HTTP, and JSONL logging of queries/requests. However auth is at best a shared secret requiring manual hardening, and request_document logs user submissions server-side (disclosed, low-risk). Deductions: weak default auth on HTTP mode.
Maintenance 9/20: MIT license explicit, v1.0.1, 1 open issue, not archived, PRs welcomed. But only 23 stars, and no evidence of release channel, security-response process, or dependency update policy. Deductions: governance and response-channel information absent.
Documentation 14/20: Thorough README: install, dual transports, tool table, config table, example output, self-disclosed limitations (stub auth, demo backend), plus an external docs site. No contradictions found statically. Deductions: no dedicated troubleshooting section; hosted docs not statically verifiable.
Setup experience 13/20: Local path is two commands (uv sync + run) with ready-made Claude Code/Cursor/Codex config examples and a zero-install hosted endpoint; env vars are well tabulated. Held under the 15 cap because no verifiable execution evidence exists.

Static review · not runListed 2026-09-08

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

Development scenarios requiring precise, citable embedded documentation facts
Users wanting quick integration and the ability to plug in custom retrieval backends

Not for

Applications needing natural-language summaries or non-verbatim answers
Scenarios where the full corpus is required but self-hosting is insufficient
Users who need authoritative source verification or dedicated support

Required permissions

Accepts queries and returns retrieval results
Can request document additions (logged only)
HTTP server and authentication on local network

Risks and side effects

Full retrieval capability is only available via the hosted service; self-hosted backend is illustrative and may not cover all documents.
The HTTP transport's bearer authentication is a stub; replace with production-grade auth before public deployment.
DNS-rebinding protection exists, but additional hardening is needed when exposing publicly.
Tool outputs come from documents that may be unofficial or outdated; users should verify facts.

Setup

Before you start

Runtime：Hosted (no local runtime) · Python 3.10+

MCP_HTTP_AUTH_TOKEN optionalsecret Optional, bearer token for HTTP transport (self-set shared-secret stub); empty means unauthenticated.

Other optional settings (7)

BYTEASK_BACKEND optional Optional, custom retrieval backend as module:callable returning a SearchBackend; empty uses the bundled SampleBackend.

BYTEASK_LOGS optional Optional, directory where query/request JSONL logs are written, defaults to logs.

MCP_TRANSPORT optional Optional, transport: stdio (local agents) or http, defaults to stdio.

MCP_HTTP_HOST optional Optional, HTTP bind host, defaults to 127.0.0.1.

MCP_HTTP_PORT optional Optional, HTTP bind port, defaults to 8000.

MCP_ALLOWED_HOSTS optional Optional, comma-separated allowed Host headers for DNS-rebinding protection in HTTP mode; * disables it.

LOG_LEVEL optional Optional, stderr log verbosity, defaults to INFO.

Ensure Python ≥3.10 and uv are installed.
Clone the repository and run uv sync.
Run uv run byteask-embedded-mcp to start the server in stdio mode.
Alternatively, connect directly to the hosted endpoint https://mcp.byteask.ai/mcp (Streamable HTTP).

claude_desktop_config.json

{
  "mcpServers": {
    "byteask-embedded-docs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.byteask.ai/mcp"
      ]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "byteask-embedded-docs": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.byteask.ai/mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add byteask-embedded-docs -- npx -y mcp-remote https://mcp.byteask.ai/mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After connecting the client to https://mcp.byteask.ai/mcp, search_docs, get_context, and request_document should appear in the tool list; calling search_docs with a known fact (e.g. Modbus function code 16) and getting a verbatim snippet with a page citation confirms the connection works.

Troubleshooting

Ensure Python version meets requirements (≥3.10) and uv is installed.
Run `uv run pytest` to check basic functionality.
Check environment variables like `BYTEASK_BACKEND`, `MCP_TRANSPORT`, `MCP_HTTP_AUTH_TOKEN`.
If using local backend, inspect logs directory (`BYTEASK_LOGS`) to debug issues.

Things to try

Once connected, you can ask your AI assistant things like:

Which Modbus function code writes multiple registers?
Look up the exact name and definition of the ETH_DMATDLAR register with a page citation
What does this SunSpec point mean exactly? Quote the source text
Request that a missing datasheet be added to the corpus

Tools 3

search_docs read-only

Search the corpus and return ranked, page-cited evidence.

get_context read-only

Expand a hit to its full source section.

request_document writes

Ask for a missing document to be added (logged server-side).

Use cases

Firmware development where exact facts like register offsets, Modbus function codes, SCPI commands are needed.

Coding agents that require verbatim snippets with page citations rather than paraphrased answers.

Experimenting with a zero-config sample backend before deciding on a self-hosted retrieval service.

Supported clients

Claude Desktop

Codex

Cursor

Listed from the project's documentation, not tested by this site.

Overview

ByteAsk Embedded MCP is an open-source MCP server that provides source-grounded, page-cited evidence retrieval for coding agents that write firmware, driver, or protocol code. It returns verbatim snippets with section and page citations, never an authored answer, and says 'no match' when nothing is relevant rather than fabricate. The server supports both stdio and Streamable HTTP transports and exposes a pluggable retrieval backend interface. This repository includes an in-memory SampleBackend for zero-setup demos; the full corpus and retrieval engine are hosted at https://mcp.byteask.ai/mcp.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision e7103adab757 Data synced 2026-10-19 Read the FMRS scoring method