On this page
Connect your Agent
Let your Agent look up what you saved in Mirabase, through MCP. You decide which spaces it can access and what it can do.
https://api.mirabase.app/mcpQuick start
Connect your MCP client
Enter the public MCP address and use Streamable HTTP transport.
Sign in and authorize
Sign in to Mirabase and choose the spaces and actions your Agent may use.
Start using it
Go back to your Agent and try finding something you saved. After you change authorization, reconnect.
Connect with a configuration file
{
"mcpServers": {
"mirabase": { "url": "https://api.mirabase.app/mcp", "transport": "streamable-http" }
}
}Connect your assistant
Every assistant uses the same address. Sign in to Mirabase, then choose the spaces it can use.
Cloud assistants
Claude (web and desktop)
Open Settings › Connectors, click Add, then Add custom connector.
Name it Mirabase and enter https://api.mirabase.app/mcp as the URL.
Follow the prompts to sign in to Mirabase and choose the spaces Claude can use.
ChatGPT
In ChatGPT on the web, open Plugins, click +, and choose Add custom MCP server. Whether you can add one depends on your ChatGPT plan and workspace settings.
Name it Mirabase and paste https://api.mirabase.app/mcp under Connection. Choose OAuth for authentication and leave Client ID and Client Secret empty.
Read the risk warning, check I understand and want to continue, then click Create as a plugin.
Install the new plugin, sign in to Mirabase, and choose the spaces ChatGPT can use. Then start a new chat, type @, and choose Mirabase.
Grok
Open grok.com/connectors, click New Connector, and choose Custom.
Enter https://api.mirabase.app/mcp as the URL.
Follow the prompts to sign in to Mirabase and choose the spaces Grok can use.
Muse Code
In your settings file, add a streamable_http entry under mcp_servers with https://api.mirabase.app/mcp as the URL.
Run the command below to sign in to Mirabase and choose spaces. The Muse app itself doesn't support custom MCP yet.
muse mcp login mirabaseLocal Agents
Claude Code
Run the command below, then type /mcp in Claude Code and choose mirabase to sign in.
claude mcp add --transport http --scope user mirabase https://api.mirabase.app/mcpCodex
Run the two commands below in order. The second opens your browser to sign in to Mirabase.
codex mcp add mirabase --url https://api.mirabase.app/mcpcodex mcp login mirabaseCursor and other editors
Add the configuration file above to its MCP settings. With Mirabase for Mac, one click on its Connect page does it.
Any other assistant that supports remote MCP (Streamable HTTP) with OAuth sign-in can connect with the same address. Assistants that only accept an API key can't connect yet.
Manage access
In the account center, you can see your Agent connections, change the spaces and actions they're allowed, or revoke access at any time.
Use your MCP client's own sign-in and authorization. Don't put long-lived credentials in its configuration.
Go to the account center →Questions
Connected, but can't find anything?
Check that the item has finished processing and that its space is authorized for this Agent. After you change authorization, reconnect and refresh the tool list.
Asked to sign in, or told you don't have permission?
401 usually means your sign-in expired. Reconnect. 403 means the current authorization doesn't allow this action. Check the spaces and tool permissions you chose.
How do I stop an Agent's access?
Revoke it under Agent connections in the account center. After that, the connection can no longer access your library.
MCP tool reference
The tools and parameters you can use are what tools/list returns after you connect. The examples below match reference version v0.6.10.
Example call
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": { "name": "mira.recall", "arguments": { "query": "When is the launch window?", "scope": { "space_ids": ["spc_product"] } } }
}mira.read
Permission: knowledge:readAvailable by defaultRead one authorized Entry. Available by default.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | ReadEntryInput | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
ReadEntryPayload{
"structuredContent": {
"entry": {
"entry_id": "example",
"space_id": "resource_01",
"tier": "knowledge",
"source_type": "example",
"title": "example",
"canonical_url": "example",
"current_version_id": "example",
"latest_version_id": "example",
"processing_status": "accepted",
"created_at": "2026-08-27T00:00:00Z",
"updated_at": "2026-08-27T00:00:00Z"
}
}
}Error responses
mira.recall
Permission: knowledge:readAvailable by defaultSearch your saved knowledge. Available by default.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | RecallRequest | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
RecallPayload{
"structuredContent": {
"results": [
{
"entry_id": "example",
"version_id": "example",
"chunk_id": "example",
"snippet": "example",
"locator": {
"type": "heading",
"heading_path": [
"example"
]
},
"source_type": "example"
}
],
"partial": false,
"trace_id": "example"
}
}Error responses
mira.memory_candidate
Permission: memory:candidateNeeds extra authorizationPropose a memory candidate without writing to a Space directly. Requires separate authorization.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | CreateMemoryCandidateRequest | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
MemoryCandidate{
"structuredContent": {
"candidate_id": "example",
"title": "example",
"content": "example",
"status": "candidate",
"created_at": "2026-08-27T00:00:00Z"
}
}Error responses
mira.confirm_memory_candidate
Permission: memory:confirmNeeds extra authorizationConfirm a memory candidate and send it through asynchronous processing. Requires separate authorization.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | ConfirmMemoryCandidateInput | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
AcceptedEntryReceipt{
"structuredContent": {
"entry_id": "example",
"version_id": "example",
"job_id": "example",
"status": "accepted"
}
}Error responses
mira.write
Permission: knowledge:writeNeeds extra authorizationSeparately authorized tool for writing knowledge.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | CreateEntryRequest | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
AcceptedEntryReceipt{
"structuredContent": {
"entry_id": "example",
"version_id": "example",
"job_id": "example",
"status": "accepted"
}
}Error responses
mira.manage
Permission: knowledge:manageNeeds extra authorizationSeparately authorized governance tool. Every organizing action requires knowledge:manage.
Parameters
Swipe sideways to see every parameter| Parameter | Type | Description |
|---|---|---|
inputRequired | McpTools.ManageInput | The input schema returned by MCP tools/list; field constraints are generated from the canonical contract. |
Response
McpTools.ManageResult{
"structuredContent": {
"entry_id": "example",
"deletion_status": "tombstoned",
"cleanup_job_id": "example"
}
}Error responses
Error codes
The HTTP status tells you the kind of error, and the error code tells you the cause. Keep the request_id for troubleshooting.
| HTTP status | Error code | Description |
|---|---|---|
| 400 | invalid_request | The request body isn't valid JSON, the Content-Type doesn't match, or a required field is missing or has the wrong type. |
| 401 | unauthorized | The Authorization header is missing, or the OIDC JWT is invalid, expired, or revoked, or its issuer or audience doesn't match. |
| 402 | quota_exceeded | The account or space has used up its quota. |
| 403 | forbidden | The caller lacks access to the target Space or Source, or the capability the operation requires. |
| 404 | not_found | The target entry_id, source_id or space_id doesn't exist, isn't accessible to the caller, or has been cleaned up (tombstoned). |
| 409 | conflict | The resource is in a conflicting state. For example, the revision doesn't match, or the entry was merged or superseded by a newer version. |
| 409 | resource_busy | The resource is being processed in the background and can't be changed right now. Try again later. |
| 413 | payload_too_large | The content or uploaded file is over the size limit for one request. Split it up, or use asynchronous upload. |
| 415 | unsupported_media_type | The file format, character encoding, or encryption can't be parsed yet. |
| 422 | validation_error | A field failed semantic validation. The field value in the response names the first invalid field. |
| 429 | rate_limited | Too many requests, or too many at once. Back off and retry as Retry-After says. Every response includes X-RateLimit-Limit / Remaining / Reset. |
| 500 | internal_error | Something went wrong inside the service. Keep the request_id from the response for troubleshooting. |
| 502 | dependency_error | An upstream service responded abnormally, so the request couldn't complete. |
| 503 | service_unavailable | The service is temporarily unavailable, for example while a model loads, an index rebuilds or a node is under maintenance. Back off and retry as Retry-After says. |
| 504 | timeout | Retrieval or answer generation timed out. Try again with a smaller depth or top_k. |
Interested in the knowledge infrastructure?
If you want to use the Mirabase Data Layer in your own product, tell us about your use case.