Codex
Add the remote server
Place this entry in your Codex configuration, runcodex mcp login feason to authorize access, then start a new session so the tool list refreshes.[mcp_servers.feason] url = "https://www.feason.com/api/mcp"
Five-minute quickstart
One question.
Inspectable context.
Connect an MCP client or make a direct REST request without installing a package. The examples use one production endpoint, one read-only credential, and only standard platform APIs.
REST endpoint
https://www.feason.com/api/context/v1/contextualizeThe path
From zero to
verified evidence.
The context is designed to be placed before a user question. Citations and coverage remain separate so your application can display and inspect them.- 01
Choose REST or MCP
Use REST when your application needs provider-neutral JSON. Use MCP when an agent or developer tool should discover and call Feason tools directly.
- 02
Create the credential
REST uses a scoped, read-only Feason Context key. MCP also requires a credential with context:read before its first request. OAuth-capable clients prompt you to sign in when connecting; private study tools require additional explicit scopes.
- 03
Make one request
Run the curl command or the dependency-free TypeScript or Python example below. Each sends the same bounded question and tradition filters.
- 04
Inspect the evidence
Read the context and citations separately. Check represented and missing traditions, provenance, token usage, and truncation before passing context to a model.
REST · curl
Create a key.
Run one command.
Sign in, create a scoped key, and keep it in a server-side environment variable. The request asks for a bounded context block while preserving tradition coverage.
- Create a scoped Context key
- The key reads only the shared canonical corpus.
- Private study context uses a separate OAuth consent flow.
Terminal
curl# Bind your key as FEASON_CONTEXT_TOKEN before running.
: "${FEASON_CONTEXT_TOKEN:?Set FEASON_CONTEXT_TOKEN first}"
curl --fail-with-body --connect-timeout 10 --max-time 20 \
https://www.feason.com/api/context/v1/contextualize \
-H "Authorization: Bearer $FEASON_CONTEXT_TOKEN" \
-H "Content-Type: application/json" \
--data '{
"query": "How did early Christians describe grace?",
"traditions": [
"Catholic",
"Eastern Orthodox",
"Reformed"
],
"maxTokens": 2000,
"maxSources": 12
}'Illustrative response shape
application/json{
"corpusVersion": "theology-v7.0.0",
"context": "# Feason Context\n...",
"citations": [{ "label": "E1", "passageId": "..." }],
"coverage": {
"representedTraditions": ["Catholic", "Eastern Orthodox"],
"missingTraditions": ["Reformed"]
},
"usage": { "estimatedTokens": 1842, "truncated": false }
}No dependency required
Use the platform you already ship.
TypeScript · Node 20+
fetchconst token = process.env.FEASON_CONTEXT_TOKEN;
if (!token?.trim()) throw new Error("Set FEASON_CONTEXT_TOKEN first");
const response = await fetch(
"https://www.feason.com/api/context/v1/contextualize",
{
method: "POST",
headers: {
Authorization: `Bearer ${token}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
"query": "How did early Christians describe grace?",
"traditions": [
"Catholic",
"Eastern Orthodox",
"Reformed"
],
"maxTokens": 2000,
"maxSources": 12
}),
signal: AbortSignal.timeout(20_000),
},
);
if (!response.ok) throw new Error(`Feason returned ${response.status}`);
const result = await response.json();
console.log(result.context);
console.log(result.citations);Python 3.11+
urllibimport json
import os
import urllib.request
payload = {
"query": "How did early Christians describe grace?",
"traditions": [
"Catholic",
"Eastern Orthodox",
"Reformed"
],
"maxTokens": 2000,
"maxSources": 12
}
request = urllib.request.Request(
"https://www.feason.com/api/context/v1/contextualize",
data=json.dumps(payload).encode(),
headers={
"Authorization": f"Bearer {os.environ['FEASON_CONTEXT_TOKEN']}",
"Content-Type": "application/json",
},
method="POST",
)
with urllib.request.urlopen(request, timeout=20) as response:
result = json.load(response)
print(result["context"])
print(result["citations"])MCP clients
Connect the agent you already use.
Claude
Add a custom connector
Enter the remote endpoint in the connectors settings, then sign in through Feason authorization to grant context:read before connecting. Private tools need additional explicit scopes.https://www.feason.com/api/mcp
ChatGPT
Create an MCP connector
Use the production endpoint for the connector URL, then approve only the scopes needed by the workflow.https://www.feason.com/api/mcp
Troubleshooting
Four checks before debugging your model.
401 Unauthorized
Confirm that FEASON_CONTEXT_TOKEN is set in the same shell and that the key has not expired or been revoked. Never place the key in browser-side code.
429 Too Many Requests
Pause requests and reduce concurrency. Honor Retry-After when present; otherwise use capped exponential backoff with jitter and a retry limit. Do not retry authentication or validation failures.
Missing tradition coverage
Treat missingTraditions as a visible corpus boundary, not proof that a tradition has no teaching on the question.
A shortened response
When usage.truncated is true, raise maxTokens within the documented limit or narrow the question and source filters.
After the first request
Build around the evidence.
Pick a workflow
Build a study companion, lesson source pack, tradition comparison, editorial review, citation checker, or focused Gospel reader.
Explore the six recipes →Keep release context
Store the returned corpusVersion with passage IDs and quotations. Verify against that release and expose incomplete checks instead of turning them into a success.
Build citation checks →Choose a client deliberately
The TypeScript and Python clients are available as source. Package manifests alone do not confirm a registry release. Follow the repository’s local build instructions and pin the source revision you use.
Explore the SDK source →