NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · Other MCP servers and tools

CVE-2026-34953: PraisonAI Has Authentication Bypass via OAuthManager.validate_token()

criticalCVSS 3.1 9.1GHSA-98f9-fqg5-hvq5

PraisonAI up to 4.5.96 accepts any Bearer token on its MCP HTTP server, because `OAuthManager.validate_token()` returns True for tokens it does not know. A network attacker can list and call every registered tool, including agent runs and container file writes. Upgrade to praisonai 4.5.97.

Published Apr 1, 2026 · Updated Nov 16, 2026 · Analysis Oct 30, 2026

What an attacker can do

In oauth.py, validate_token() loops over its stored tokens and returns the expiry check if one matches. If none matches, the function falls through to return True. The token store is empty by default, so every token passes, including fake_token_abc123.

The MCP server started by praisonai mcp serve --transport http-stream binds to 0.0.0.0 by default with no API key required. An unauthenticated attacker who can reach the port sends any Authorization: Bearer value and gets the full tool list. The advisory's proof of concept returns 200 OK with more than 50 tools, among them praisonai.agent.run, praisonai.workflow.run and praisonai.containers.file_write.

That gives the attacker agent execution, workflow runs, container file read and write, and skill loading. In practice that is remote control of an agent that holds your model API keys and tool credentials, and possibly code execution through file writes into containers. The CVSS vector rates confidentiality and integrity as high and availability as none. Any deployment of the HTTP MCP server on a shared network or the internet is exposed.

Who is affected

praisonai (pip) up to and including 4.5.96, when the MCP server runs over HTTP (--transport http-stream). Check:

pip show praisonai

Check whether a praisonai mcp serve process is running, and whether its port answers from another host.

praisonai

Ecosystem
PyPI
Affected
<= 4.5.96
Fixed in
4.5.97
Check
pip show praisonai

How to fix it

Upgrade to 4.5.97:

pip install --upgrade praisonai==4.5.97

The advisory's suggested fix makes validate_token() return False for unknown tokens, with an introspection call for external or JWT tokens. The sources give no further workaround. Until you upgrade, bind the server to 127.0.0.1 or put it behind an authenticating reverse proxy, which is general hardening rather than something the advisory states. Review what the agent could have done while exposed.

How to tell if you were hit

Look in the MCP server's access logs for requests to /mcp from unfamiliar addresses with Bearer tokens you never issued, especially tools/list followed by tools/call for praisonai.agent.run, praisonai.workflow.run or praisonai.containers.file_write. Check containers for files you did not create. Run a probe: send the PoC request with Bearer fake_token_abc123 to your own server. A 200 with a tool list means it is vulnerable.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Summary

OAuthManager.validate_token() returns True for any token not found in its internal store, which is empty by default. Any HTTP request to the MCP server with an arbitrary Bearer token is treated as authenticated, granting full access to all registered tools and agent capabilities.

Details

oauth.py:364 (source) -> oauth.py:374 (loop miss) -> oauth.py:381 (sink)

# source
def validate_token(self, token: str) -> bool:
    for stored_token in self._tokens.values():
        if stored_token.access_token == token:
            return not stored_token.is_expired()

# sink -- _tokens is empty by default, loop never executes, falls through
    return True

PoC

# install: pip install -e src/praisonai
# start server: praisonai mcp serve --transport http-stream --port 8080

curl -s -X POST http://127.0.0.1:8080/mcp \
  -H "Authorization: Bearer fake_token_abc123" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":1}'

# expected output: 200 OK with full tool list (50+ tools)
# including praisonai.agent.run, praisonai.workflow.run, praisonai.containers.file_write

Impact

Any unauthenticated attacker with network access to the MCP HTTP server can call all registered tools including agent execution, workflow runs, container file read/write, and skill loading. The server binds to 0.0.0.0 by default with no API key required.

Suggested Fix

def validate_token(self, token: str) -> bool:
    for stored_token in self._tokens.values():
        if stored_token.access_token == token:
            return not stored_token.is_expired()
    # Unknown tokens must be rejected.
    # For external/JWT tokens, call the introspection endpoint here before returning.
    return False

Credited to YeranG30.

Questions

Does the attacker need a token?

No. Any string works as a Bearer token, because unknown tokens are treated as valid. The token store is empty by default.

Is the stdio MCP transport affected?

The advisory describes only the HTTP server (http-stream). The sources do not discuss stdio.

Which version fixes it?

praisonai 4.5.97. Versions up to 4.5.96 are vulnerable.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More Other MCP servers and tools vulnerabilities

All Other MCP servers and tools entries