context7 MCP Docker image fails: "Server exited before responding to initialize request"

Question

When connecting the context7 MCP server to VS Code through the Docker MCP Toolkit, the server fails to start: the log shows the container reports it is running on HTTP at localhost:8080/mcp, then exits with "Server exited before responding to initialize request." Other Docker-based MCP servers work fine. Why does this one fail and what configuration makes it work?

Originally asked on GitHub: https://github.com/upstash/context7/issues/300

AgentArmor team · Accepted Answer

**Status: explained.** The failure isn't a config mistake on your side: the `mcp/context7` Docker image is preconfigured to start Context7 over HTTP/SSE, while the Docker MCP Toolkit launches stdio processes, so the server exits before it can answer the `initialize` request. A maintainer confirmed the image is not an official Context7 image (an official one is planned) and that the server normally defaults to stdio when no transport is selected — but the Dockerfile explicitly enables HTTP because Smithery deploys from that Dockerfile.

The working workaround, confirmed by another user in the thread, is to force stdio via an environment variable:

```json
"context7": {
  "command": "docker",
  "args": ["run", "-i", "--rm", "-e", "MCP_TRANSPORT=stdio", "mcp/context7"]
}
```

Do **not** add `--transport stdio` to the container args: the reporter tried that and got `node: bad option:-- transport` with exit code 9, because the image entrypoint already sets a transport.

If your client supports HTTP MCP, the remote endpoint `https://mcp.context7.com/mcp` also works: `"type": "http"` in VS Code and `"type": "streamable-http"` in Roo Code. The reporter found the remote endpoint failed in Cline, though.

Note: per the maintainer, `mcp/context7` on Docker Hub is not an official image; at the time of the thread an update to the Dockerfile to default to stdio was pending on Smithery's side.

AgentArmor team · Answer

**What a GitHub user tried on GitHub** (Didn't work)

Reporter used the standard Docker MCP Toolkit config in VS Code: `docker run -i --rm mcp/context7`. The container logged that it was running on HTTP at localhost:8080/mcp, then exited before responding to `initialize`.

Source: https://github.com/upstash/context7/issues/300

AgentArmor team · Answer

**What @ro1111bot tried on GitHub** (Worked)

A commenter suggested forcing stdio with the environment variable `MCP_TRANSPORT=stdio` via `-e` in the docker run args. Another user later confirmed this fixes the connection problem.

Source: https://github.com/upstash/context7/issues/300#issuecomment-3024173947

AgentArmor team · Answer

**What @enesgules tried on GitHub** (Didn't work)

A maintainer suggested adding `--transport stdio` to the docker args. The reporter tested it and got `node: bad option:-- transport` with exit code 9, because the image already ships with a transport configured.

Source: https://github.com/upstash/context7/issues/300#issuecomment-2983026645

AgentArmor team · Answer

**What @KikeVen tried on GitHub** (Partly worked)

Reporter used the remote HTTP endpoint `https://mcp.context7.com/mcp` with `"type": "http"` in VS Code and found the `"type": "streamable-http"` setting for Roo Code. It worked in VS Code, but Cline could not use that URL setting.

Source: https://github.com/upstash/context7/issues/300#issuecomment-2984683714

AgentArmor team · Answer

**What @kabakaev tried on GitHub** (Outcome unclear)

A commenter suggested overriding the image CMD so the container runs `node dist/index.js --transport stdio` directly with a custom image name. This was posted as a suggested config but not confirmed as working in the thread.

Source: https://github.com/upstash/context7/issues/300#issuecomment-3124555983

context7 MCP Docker image fails: "Server exited before responding to initialize request"

6 answers

Add your answer or comment

context7 MCP Docker image fails: "Server exited before responding to initialize request"

6 answers

Add your answer or comment

Related questions