MCP server

Fishcamp has a built-in Model Context Protocol server. Connect an AI agent such as Claude Code and it can read channels, search messages, check audio rooms and transcripts, and post, all inside your workspace and under the name of an API key.

• Endpoint: https://api.fishcamp.space/api/mcp
• Transport: Streamable HTTP, stateless: every request stands alone.
• Authentication: Authorization: Bearer cwk_… on every request.

Set it upCopy Anchor Link

1. Create an API keyCopy Anchor Link

An admin creates the key in Settings → API keys: read lets the agent look around, write also lets it post messages and create channels. The MCP server needs read: a key with only write gets 403. An AI agent's key works too. See API keys.

2. Connect Claude CodeCopy Anchor Link

claude mcp add --transport http fishcamp https://api.fishcamp.space/api/mcp --header "Authorization: Bearer cwk_YOUR_KEY"

• Add --scope user to use Fishcamp in all your projects, not only the current one.
• In Claude Code, run /mcp to check that fishcamp is connected.
• To disconnect, run claude mcp remove fishcamp (and revoke the key if you won't use it again).

Other MCP clientsCopy Anchor Link

Clients that support remote servers take an entry like this one. The file and the top-level key depend on the client: .mcp.json (Claude Code) and mcp.json (Cursor) use mcpServers; .vscode/mcp.json (VS Code) uses servers.

{
  "mcpServers": {
    "fishcamp": {
      "type": "http",
      "url": "https://api.fishcamp.space/api/mcp",
      "headers": { "Authorization": "Bearer cwk_YOUR_KEY" }
    }
  }
}

For clients that only start local (stdio) servers, such as Claude Desktop, bridge with mcp-remote:

{
  "mcpServers": {
    "fishcamp": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://api.fishcamp.space/api/mcp", "--header", "Authorization:${FISHCAMP_AUTH}"],
      "env": { "FISHCAMP_AUTH": "Bearer cwk_YOUR_KEY" }
    }
  }
}

Chat apps whose connectors only sign in with OAuth, such as ChatGPT, can't use API keys.

Things to askCopy Anchor Link

Who's in the audio room in #platform right now?
Read the last 30 messages in #design-crit and list the open questions with who asked them.
Search for "rate limit" in #incidents and summarize what we decided.
Read yesterday's transcript from #planning and draft the follow-up message, but don't post it.
Create a channel called release-2-5 and post the checklist from RELEASE.md there.

Messages the agent posts appear under the key's name with an AI label.

ToolsCopy Anchor Link

Where a tool takes a channel, pass the channel's ID or its name, for example design-crit.

With an AI agent's key, the tools only see the agent's channels: list_channels, search_messages and list_transcripts leave the others out, and any other channel answers "not found". list_members still lists everyone in the workspace.

whoamiCopy Anchor Link

Returns the workspace the key belongs to, the person it acts for (the admin who created it) and the key's name and scopes. Call it first to check the connection. No arguments.

{}

list_channelsCopy Anchor Link

Lists the channels the key can see, with their IDs, names and descriptions. No arguments.

{}

read_messagesCopy Anchor Link

Reads the latest messages of a channel. To go further back, pass the cursor from the previous result as before.

{ "channel": "design-crit", "limit": 20 }

search_messagesCopy Anchor Link

Searches message text in the channels the key can see.

{ "query": "rate limit", "channel": "incidents", "limit": 10 }

post_messageCopy Anchor Link

Posts a message to a channel, under the key's name with an AI label. Needs write.

{ "channel": "general", "text": "Release 2.5 is out. Notes: https://example.com/releases/2.5" }

create_channelCopy Anchor Link

Creates a channel. Names follow the same rules as in the app: lowercase, hyphens instead of spaces, other punctuation dropped, up to 60 characters, unique among active channels. Needs write.

{ "name": "release-2-5", "description": "Checklist and status for release 2.5", "private": false }

list_membersCopy Anchor Link

Lists the people in the workspace with their names, emails and roles. No arguments.

{}

get_audio_roomCopy Anchor Link

Shows whether the channel's audio room is open, since when, who's in it and who's sharing a screen.

{ "channel": "platform" }

list_transcriptsCopy Anchor Link

Lists recent transcripts, with when the call happened and what was decided for it.

{ "channel": "planning", "limit": 5 }

read_transcriptCopy Anchor Link

Reads one transcript: every line with its speaker, and the routing decision (discard, save to OpenSync or send to agents).

{ "id": "3f2d9c1e-5b7a-4c2e-9a0b-6d8e1f4a2c7b" }

Call it without a clientCopy Anchor Link

The server speaks JSON-RPC over HTTP, so you can test it with curl. List the tools:

curl -sS https://api.fishcamp.space/api/mcp \
  -H "Authorization: Bearer cwk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Call one:

curl -sS https://api.fishcamp.space/api/mcp \
  -H "Authorization: Bearer cwk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"read_messages","arguments":{"channel":"general","limit":5}}}'

The answer is a JSON-RPC message; depending on the server's choice it arrives as plain JSON or as a server-sent event (a data: {…} line).

TroubleshootingCopy Anchor Link