client must accept both application/json and text/event-stream error mcp

Not specified by the original author.

Fix Tavily MCP Not Acceptable: Client must accept both application/json and text/event-stream

TL;DR

Switch from the type: http config to the stdio config (npx -y tavily-mcp@latest with your key in env). The hosted Tavily MCP endpoint requires an Accept: application/json, text/event-stream header that Claude Code's HTTP client does not send, so every call fails with what looks like an authentication error.

What the failure looks like:

{"error":"Search failed","detail":"Authentication required"}

and at the raw HTTP level:

Not Acceptable: Client must accept both application/json and text/event-stream

Steps

1. Confirm the header is the problem

Run a raw request without the special Accept header:

curl -X POST "https://mcp.tavily.com/mcp/?tavilyApiKey [your value] Tavily API key]" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

Success check: you get the Not Acceptable response, which proves the header (not your key) is the issue. With the right header it works:

curl -X POST "https://mcp.tavily.com/mcp/?tavilyApiKey [your value] Tavily API key]" \
  -H "Accept: application/json, text/event-stream" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

2. Replace the HTTP config with stdio

In your MCP config, drop the type: http entry and use:

{
  "mcpServers": {
    "tavily": {
      "command": "npx",
      "args": ["-y", "tavily-mcp@latest"],
      "env": { "TAVILY_API_KEY": "[your Tavily API key]" }
    }
  }
}

Success check: restart the client; tavily-search and tavily-extract appear and searches return results.

3. Validate the key separately if it still fails

If the stdio route also fails, verify the key itself:

curl -s "https://api.tavily.com/search" -H "Content-Type: application/json" \
  -d '{"api_key":"[your Tavily API key]","query":"test","max_results":1}' | head -c 300

Success check: valid JSON with results means the key is fine and the problem is the transport.

When this applies

• You configured Tavily with "type": "http" and `"url": "https://mcp.tavily.com/mcp/?tavilyApiKey [your value] and every tool call returns an authentication error.
• Raw curl shows the Not Acceptable message about text/event-stream.

When it does not apply

• The stdio npx tavily-mcp server fails to start at all (check Node and npx first).
• The key is actually invalid or missing; the API check in step 3 returns a 401. Regenerate the key instead.

Tool compatibility

• tavily-mcp npm package (latest)
• Tavily hosted MCP at mcp.tavily.com
• Claude Code (the type: http client is the one affected), also relevant to any client that cannot set custom Accept headers

Why it happens

The Tavily MCP server streams responses over SSE, so it requires clients to send Accept: application/json, text/event-stream. Claude Code's type: http client sends a default Accept header without the SSE part, the server rejects the request as Not Acceptable, and the client surfaces that as an authentication failure, which sends you chasing the API key instead of the header.

Edge cases

• Do not paste the key with quotes or trailing spaces in the URL parameter; the type: http setup passes it verbatim in the query string.
• If you need the remote endpoint (no local node), use mcp-remote with the URL as an npx stdio wrapper instead of type: http.
• API rate limits on the free tier return HTTP 429, not auth errors; treat those as usage limits, not this problem.

A user configured the Tavily hosted MCP server with Claude Code type http pointing at https://mcp.tavily.com/mcp/ with a tavilyApiKey query parameter. Every tool call failed with {"error":"Search failed","detail":"Authentication required"}. The issue thread diagnosed that the Tavily MCP server requires clients to send the header Accept: application/json, text/event-stream, which Claude Code's type http client does not include by default, and that curl without the header returns Not Acceptable: Client must accept both application/json and text/event-stream. The verified workaround is the stdio-based config (npx -y tavily-mcp@latest with TAVILYAPIKEY in env) or mcp-remote. Source: https://github.com/tavily-ai/tavily-mcp/issues/121

Not specified by the original author.

TL;DR: Transport choice is the decision that matters. import { createMCPClient } from '@ai-sdk/mcp'; const mcpClient = await createMCPClient({ transport: { type: 'http', url: 'YOUR_HOST.com/mcp', headers: { Authorization: 'your auth header' }, }, }); const tools = await mcpClient.tools(); // pass tools into streamText / ToolLoopAgent like any other tools Rules: 1.

The AI SDK connects to MCP servers for tools, resources, and prompts through @ai-sdk/mcp. Transport choice is the decision that matters.

import { createMCPClient } from '@ai-sdk/mcp';

const mcpClient = await createMCPClient({ transport: { type: 'http', url: 'YOUR_HOST.com/mcp', headers: { Authorization: 'your auth header' }, }, });

const tools = await mcpClient.tools(); // pass tools into streamText / ToolLoopAgent like any other tools

Rules:

1. Use HTTP transport (StreamableHTTPClientTransport style) for production. It is the documented recommendation.
2. stdio transport is for local development only. It cannot be deployed to production runtimes (no local subprocess on serverless/edge).
3. SSE transport exists as an alternative HTTP option. Prefer the plain HTTP transport unless the server only speaks SSE.
4. The client handles MCP protocol negotiation, including newer stateless versions with fallback to the legacy initialize handshake. You do not hand-roll the handshake.
5. If you use OpenAI's Responses API, there is a built-in openai.tools.mcp option that integrates MCP server-side without converting tools client-side. Check the OpenAI provider docs for it.
6. Treat MCP tools like untrusted third-party tools: validate their outputs before rendering or acting on them, and gate destructive ones with toolApproval.
7. Close the client when done (mcpClient.close()) in long-lived processes to avoid leaking connections.

Verification: list the tools after connect and confirm names and schemas look right before letting the model call them. A misconfigured transport usually fails here, not at call time.

When to use

You hit exactly this: MCP tools in AI SDK: HTTP transport, client setup, and production rules in MCP tools in AI SDK.

When not to use

A different error, or the same symptom in a different tool. This page only covers the failure above.

Compatibility

MCP tools in AI SDK.

Verified against the live AI SDK docs 2026-09-26. The MCP tools guide documents createMCPClient with HTTP transport as the production recommendation, stdio as local-only, SSE as an alternative, protocol version negotiation, and the OpenAI Responses API built-in MCP tool option.

Not specified by the original author.

TL;DR

Issue strands-agents/harness-sdk#482 (closed, 14 comments): # Overview Add support for loading MCP server configurations from a JSON config file, allowing users to define multiple MCP servers declaratively instead of programmatically.

Steps

1. Issue strands-agents/harness-sdk#482 (closed, 14 comments): # Overview Add support for loading MCP server configurations from a JSON config file, allowing users to define multiple MCP servers declaratively instead of programmatically. This extends the existing experimental agent_config.py to support an mcp_servers field that creates and manages MCPClient instances automatically. Note: Multiple MCP servers and automatic lifecycle management already work programmatically (see tests_integ/mcp/test_mcp_tool_provider.py::test_mcp_client_multiple_servers). This feature request is specifically about configuration file support.
2. [yonib05 (maintainer)]: Shipped: you can now load multiple MCP servers from a config file with load_servers instead of wiring up MCPClient instances by hand. It is in both SDKs (Python PR #3053, TypeScript PR #2947). The config uses the mcpServers JSON format with env-var interpolation, a disabled flag per server, and per-server error tolerance. Docs: mount the host path https at container path //strandsagents.com/latest/documentation/docs/user-guide/concepts/tools/mcp-tools/

When to use

You are seeing this: This extends the existing experimental agent\config.py to support an mcp\servers field that creates and manages MCPClient instances automatically. Use this skill when you run into "[FEATURE] Support for multiple MCP servers (and loading from config file)".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

Edge cases

• Note: Multiple MCP servers and automatic lifecycle management already work programmatically (see tests\integ/mcp/test\mcp\tool\provider.py::test\mcp\client\multiple\servers).

In gh:strands-agents/harness-sdk#482 (AWS Strands Agents), the issue reported was: Issue strands-agents/harness-sdk#482 (closed, 14 comments): # Overview Add support for loading MCP server configurations from a JSON config file, allowing users to define multiple MCP servers declaratively instead of programmatically. This extends the existing experimental agent_config.py to support an mcp_servers field that creates and manages MCPClient instances automatically. Note: Multiple MCP servers and automatic lifecycle management already work programmatically (see tests_integ/mcp/test_mcp_tool_provider.py::test_mcp_client_multiple_servers). This feature request is specifically about configuration file support. The verified resolution from the thread: [yonib05 (maintainer)]: Shipped: you can now load multiple MCP servers from a config file with load_servers instead of wiring up MCPClient instances by hand. It is in both SDKs (Python PR #3053, TypeScript PR #2947). The config uses the mcpServers JSON format with env-var interpolation, a disabled flag per server, and per-server error tolerance. Docs: mount the host path https at container path //strandsagents.com/latest/documentation/docs/user-guide/concepts/tools/mcp-tools/ Source: mount the host path https at container path //github.com/strands-agents/harness-sdk/issues/482