CalvinBackup/invariant-gateway
1
0
Fork 0
mirror of https://github.com/invariantlabs-ai/invariant-gateway.git synced 2026-08-11 19:40:21 +02:00
Code Issues Packages Projects Releases Wiki Activity

Address comments on PR and update README.

Browse Source
This commit is contained in:
Hemang
2025-06-04 11:46:35 +02:00
committed by Hemang Sarkar
parent cc3e96c20a
commit 05e09331e9
5 changed files with 60 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+27 -3
View File
@@ -279,22 +279,44 @@ export ANTHROPIC_API_KEY={your-anthropic-api-key};invariant-auth={your-invariant
This setup ensures that SWE-agent works seamlessly with Invariant Gateway, maintaining compatibility while enabling full functionality. 🚀
### **Using MCP with Invariant Gateway**
Invariant Gateway supports MCP (both stdio and SSE transports) tool calling.
Invariant Gateway supports MCP (stdio, SSE and Streamable http) tool calling.
For stdio transport based MCP, follow steps [here](https://github.com/invariantlabs-ai/invariant-gateway/tree/main/gateway/mcp).
For SSE transport based MCP, here are the steps to point your MCP client to a local instance of the Invariant Gateway which will then proxy all calls to the MCP server:
For **SSE transport based MCP**, here are the steps to point your MCP client to a local instance of the Invariant Gateway which will then proxy all calls to the MCP server while guardrailing:
* Run the Gateway locally by following the steps [here](https://github.com/invariantlabs-ai/invariant-gateway/tree/main?tab=readme-ov-file#run-the-gateway-locally).
* Use the following configuration to connect to the local Gateway instance:
```python
await client.connect_to_sse_server(
from mcp.client.sse import sse_client
await connect_to_sse_server(
server_url="http://localhost:8005/api/v1/gateway/mcp/sse",
headers={
"MCP-SERVER-BASE-URL": "<The base URL to your MCP server>",
"INVARIANT-PROJECT-NAME": "<The Invariant dataset name>",
"PUSH-INVARIANT-EXPLORER": "true",
"INVARIANT-API-KEY": "<your-invariant-api-key>"
"INVARIANT-X-MCP-SERVER-{CUSTOM-MCP-SERVER-HEADER-NAME}": "<custom-value-passed-to-mcp-server>"
},
)
```
For **Streamable HTTP transport based MCP**, here are the steps to point your MCP client to a local instance of the Invariant Gateway which will then proxy all calls to the MCP server while guardrailing:
* Run the Gateway locally by following the steps [here](https://github.com/invariantlabs-ai/invariant-gateway/tree/main?tab=readme-ov-file#run-the-gateway-locally).
* Use the following configuration to connect to the local Gateway instance:
```python
from mcp.client.streamable_http import streamablehttp_client
await streamablehttp_client(
url="http://localhost:8005/api/v1/gateway/mcp/sse",
headers={
"MCP-SERVER-BASE-URL": "<The base URL to your MCP server>",
"INVARIANT-PROJECT-NAME": "<The Invariant dataset name>",
"PUSH-INVARIANT-EXPLORER": "true",
"INVARIANT-API-KEY": "<your-invariant-api-key>"
"INVARIANT-X-MCP-SERVER-{CUSTOM-MCP-SERVER-HEADER-NAME}": "<custom-value-passed-to-mcp-server>"
},
)
```
@@ -303,6 +325,8 @@ The `INVARIANT-API-KEY` header is used both for pushing the traces to explorer a
If no `INVARIANT-PROJECT-NAME` header is specified but `PUSH-INVARIANT-EXPLORER` is set to "true", a new Invariant project will be created and the MCP traces will be pushed there.
If you pass a header called `INVARIANT-X-MCP-SERVER-CUSTOM-API-KEY`, it will be passed as the `CUSTOM-API-KEY` header to the underlying MCP server.
You can also specify blocking or logging guardrails for the project name by visiting the Explorer.
---